[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"cms-page:\u002Fabout\u002Four-team\u002Fharry-brace\u002Fpage\u002F2":3},{"name":4,"url":5,"description":6,"avatar_urls":7,"acf":9,"social":195,"head_links":197,"breadcrumbs":183,"posts":201,"pages":178,"head":1475},"Harry Brace","","I'm the Director, Head of Solutions at Impression. I spend my time debugging and implementing tracking using Google Tag Manager, managing data warehousing platforms and building Looker Studio reports!",{"24":8,"48":8,"96":8},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress&fit=fit&fm=jpg&h=96&q=90&w=96",{"full_bio":10,"job_title":11,"specialist_service_links":12,"photo":49,"page_content_modules":105,"hero":179},"\u003Cp>Director, Head of Solutions. Joined Impression in November 2017.\u003C\u002Fp>\n\u003Cp>I gained 4 years of experience as a web developer before moving to Impression as a digital account manager in 2017. I moved across to the Analytics department in 2019 and I’ve been strengthening and growing our Analytics offering ever since.\u003C\u002Fp>\n\u003Cp>Day to day, I spend my time debugging and implementing tracking using Google Tag Manager, managing data warehousing platforms and building Looker Studio reports!\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Something that I’m proud of:\u003C\u002Fstrong>\u003Cbr \u002F>\nBuilding the analytics department over the last couple of years with the help of Aaron and now Alex H.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Outside of work:\u003C\u002Fstrong>\u003Cbr \u002F>\nFind me socialising, skiing or gaming.\u003C\u002Fp>\n","Director, Head of Solutions",[13,30,40],{"page":14,"link_text":29},{"ID":15,"post_author":16,"post_date":17,"post_date_gmt":17,"post_content":5,"post_title":18,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":21,"to_ping":5,"pinged":5,"post_modified":22,"post_modified_gmt":22,"post_content_filtered":5,"post_parent":23,"guid":24,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":28},374,"98","2021-01-11 16:25:32","Analytics","publish","closed","analytics","2026-09-28 11:36:16",0,"\u002F?page_id=374","page","0","raw","\u002Fanalytics\u002F","Web Analytics",{"page":31,"link_text":34},{"ID":32,"post_author":16,"post_date":33,"post_date_gmt":33,"post_content":5,"post_title":34,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":35,"to_ping":5,"pinged":5,"post_modified":36,"post_modified_gmt":36,"post_content_filtered":5,"post_parent":37,"guid":38,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":39},17916,"2022-06-20 08:56:55","Tag Management","tag-management","2026-09-28 11:05:27",17552,"\u002F?page_id=17916","\u002Fmedia-solutions\u002Ftag-management\u002F",{"page":41,"link_text":44},{"ID":42,"post_author":16,"post_date":43,"post_date_gmt":43,"post_content":5,"post_title":44,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":45,"to_ping":5,"pinged":5,"post_modified":46,"post_modified_gmt":46,"post_content_filtered":5,"post_parent":37,"guid":47,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":48},17140,"2022-05-27 14:22:19","Google Marketing Platform","google-marketing-platform","2026-09-25 10:43:13","\u002F?page_id=17140","\u002Fmedia-solutions\u002Fgoogle-marketing-platform\u002F",{"ID":50,"id":50,"title":51,"filename":52,"filesize":53,"url":54,"link":55,"alt":5,"author":56,"description":5,"caption":5,"name":57,"status":58,"uploaded_to":23,"date":59,"modified":59,"menu_order":23,"mime_type":60,"type":61,"subtype":62,"icon":63,"width":64,"height":65,"sizes":66},25135,"Harry-Brace2","Harry-Brace2.png",306919,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&q=90","\u002Fharry-brace2\u002F","12","harry-brace2","inherit","2023-10-03 16:03:03","image\u002Fpng","image","png","\u002Fwp\u002Fwp-includes\u002Fimages\u002Fmedia\u002Fdefault.png",696,600,{"thumbnail":67,"thumbnail-width":68,"thumbnail-height":68,"medium":69,"medium-width":70,"medium-height":71,"medium_large":72,"medium_large-width":73,"medium_large-height":74,"large":75,"large-width":76,"large-height":77,"1536x1536":78,"1536x1536-width":79,"1536x1536-height":80,"2048x2048":81,"2048x2048-width":82,"2048x2048-height":83,"post-thumbnail":84,"post-thumbnail-width":85,"post-thumbnail-height":86,"12_col":87,"12_col-width":88,"12_col-height":89,"default_landscape":90,"default_landscape-width":91,"default_landscape-height":65,"default_portrait":92,"default_portrait-width":65,"default_portrait-height":91,"6_col_and_margin_short":93,"6_col_and_margin_short-width":85,"6_col_and_margin_short-height":89,"11_col_and_margin":94,"11_col_and_margin-width":95,"11_col_and_margin-height":89,"9_col_rectangle":96,"9_col_rectangle-width":97,"9_col_rectangle-height":98,"11_col":99,"11_col-width":100,"11_col-height":89,"staff_image_696_600":101,"staff_image_696_600-width":64,"staff_image_696_600-height":65,"rectangular_thumbnail":102,"rectangular_thumbnail-width":103,"rectangular_thumbnail-height":104},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=640&q=90&w=640","640","https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=scale&h=172&q=90&w=200",200,172,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=scale&h=662&q=90&w=768",768,662,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=1000&q=90&w=1892","1892","1000","https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=scale&h=1324&q=90&w=1536",1536,1324,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=scale&h=1766&q=90&w=2048",2048,1766,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=1180&q=90&w=1860",1860,1180,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2880",2880,870,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=600&q=90&w=800",800,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=800&q=90&w=600","https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=870&q=90&w=1860","https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=870&q=90&w=3080",3080,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=1400&q=90&w=2150",2150,1400,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2640",2640,"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=600&q=90&w=696","https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress%2Cformat&fit=crop&h=300&q=90&w=750",750,300,[106],{"acf_fc_layout":107,"reduced":108,"top_overlap":-1,"bottom_underlap":-1,"disable_observer":-1,"title":109,"posts":110},"author_posts",true,"Harry Brace’s posts",{"data":111,"pages":178},[112,126,137,149,159,168],{"type_key":113,"title":114,"excerpt":115,"to":116,"postedAt":117,"readingTime":118,"image":119,"authors":120,"category":123},"post","Google Meridian: What you need to know","\u003Cp>Google Meridian launched in March 2024, offering digital marketers exciting new ways to explore how their campaigns are influencing sales and other business-critical goals. It’s an open-source Media Mix Modelling (MMM) platform that empowers you to overcome common marketing effectiveness measurement challenges. For context, MMM is a cookieless form of measurement which utilises statistical techniques […]\u003C\u002Fp>\n","\u002Fblog\u002Fgoogle-meridian-what-you-need-to-know\u002F","2025-01-17T09:13:55",7,"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=640&q=90&w=640",[121],{"name":4,"image":122},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress&fit=crop&fm=jpg&h=150&q=80&w=150&bg=CFE6F9",{"to":124,"name":125},"\u002Fblog\u002Fcategory\u002Fdigital-marketing\u002F","Digital Marketing",{"type_key":113,"title":127,"excerpt":128,"to":129,"postedAt":130,"readingTime":118,"image":131,"authors":132,"category":134},"Server-side tagging in Google Tag Manager (GTM)","\u003Cp>Server-side tagging is now available within Google Tag Manager (GTM), putting better marketing effectiveness data back into your hands. Packed with tips from our in-house experts, this guide unpacks the principles of server-side tagging and how you can use it to evolve your measurement strategy in GTM. Read on to discover the benefits of this […]\u003C\u002Fp>\n","\u002Fblog\u002Fserver-side-tagging-gtm\u002F","2025-03-31T14:07:58","https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=640&q=90&w=640",[133],{"name":4,"image":122},{"to":135,"name":136},"\u002Fblog\u002Fcategory\u002Fmeasurement\u002F","Measurement",{"type_key":113,"title":138,"excerpt":139,"to":140,"postedAt":141,"readingTime":142,"image":143,"authors":144,"category":146},"Understanding the workings of Bayesian A\u002FB testing","\u003Cp>In digital marketing, A\u002FB testing is a technique for comparing two versions of a variable, such as a landing page layout or ad creative, to determine which version performs best with the target audience. The audience is randomly divided into two groups: one group is exposed to version A (the control), and the other to […]\u003C\u002Fp>\n","\u002Fblog\u002Funderstanding-the-workings-of-bayesian-ab-testing\u002F","2024-12-19T17:20:34",10,"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=640&q=90&w=640",[145],{"name":4,"image":122},{"to":147,"name":148},"\u002Fblog\u002Fcategory\u002Fdata-engineering\u002F","Data Engineering",{"type_key":113,"title":150,"excerpt":151,"to":152,"postedAt":153,"readingTime":154,"image":155,"authors":156,"category":158},"Balancing granularity in Marketing Mix Modelling","\u003Cp>Marketing Mix Modeling (MMM) is a statistical method used to measure the effectiveness of marketing activities and other factors on sales, helping you understand how your marketing efforts contribute to results – with a holistic methodology. MMM enables you to optimise budget allocation by analysing the effectiveness of each marketing channel and identifying the point […]\u003C\u002Fp>\n","\u002Fblog\u002Fgranularity-marketing-mix-modelling\u002F","2024-10-03T16:37:11",8,"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=640&q=90&rect=0%2C1200%2C2400%2C2400&w=640",[157],{"name":4,"image":122},{"to":147,"name":148},{"type_key":113,"title":160,"excerpt":161,"to":162,"postedAt":163,"readingTime":142,"image":164,"authors":165,"category":167},"A deep dive into Google Big Query ML for data-driven marketers","\u003Cp>Google BigQuery ML (BQML) is a powerful tool that allows data scientists and analysts to build and deploy machine learning models directly within Google BigQuery. By integrating machine learning into BigQuery, BQML eliminates the need to move data between various systems or platforms, enabling you to utilise data science to extract data-driven insights efficiently. This […]\u003C\u002Fp>\n","\u002Fblog\u002Fgoogle-big-query-ml\u002F","2026-08-25T09:29:44","https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=640&q=90&w=640",[166],{"name":4,"image":122},{"to":147,"name":148},{"type_key":113,"title":169,"excerpt":170,"to":171,"postedAt":172,"readingTime":173,"image":174,"authors":175,"category":177},"Building trustworthy media insights through model validation","\u003Cp>What is model validation and why is it important? Model validation is the process of evaluating a statistical model’s performance and confirming the model achieves its intended purpose. It involves a series of techniques and assessments to determine how well a model predicts outcomes using new data that were not part of the training set. […]\u003C\u002Fp>\n","\u002Fblog\u002Fbuilding-trustworthy-media-insights-through-model-validation\u002F","2024-06-24T15:44:16",5,"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=640&q=90&w=640",[176],{"name":4,"image":122},{"to":124,"name":125},"6",{"type":180,"style":181,"title":182,"heading":4,"intro":10,"services":12,"breadcrumbs":183,"photo":67,"social":195},"author","dark","Meet the Team",[184,187,190,193],{"path":185,"name":186},"\u002F","Home",{"path":188,"name":189},"\u002Fabout\u002F","About Us",{"path":191,"name":192},"\u002Fabout\u002Four-team\u002F","Meet the team",{"path":194,"name":4},null,{"twitter":5,"linkedin":196},"https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fharrison-brace-137b8482\u002F",[198],{"rel":199,"href":200},"canonical","https:\u002F\u002Fwww.impressiondigital.com\u002Fabout\u002Four-team\u002Fharry-brace\u002F",[202,527,716,908,1097,1287],{"id":203,"date":204,"date_gmt":204,"guid":205,"modified":117,"modified_gmt":117,"slug":207,"status":19,"type":113,"link":116,"title":208,"content":209,"excerpt":212,"author":213,"featured_media":214,"comment_status":20,"ping_status":20,"sticky":211,"template":5,"format":215,"meta":216,"categories":218,"collections":221,"coauthors":222,"class_list":224,"acf":233,"head_links":234,"breadcrumbs":235,"url_path":116,"search_description":217,"featured_image_urls":242,"authors":277,"type_nicename":290,"reading_minutes":118,"category":291,"yoast_title":114,"yoast_meta":294,"yoast_json_ld":350,"_links":477},27600,"2025-01-17T09:00:18",{"rendered":206},"\u002F?p=27600","google-meridian-what-you-need-to-know",{"rendered":114},{"rendered":210,"protected":211},"\n\u003Cp>Google Meridian launched in March 2024, offering digital marketers exciting new ways to explore how their campaigns are influencing sales and other business-critical goals. It’s an open-source Media Mix Modelling (MMM) platform that empowers you to overcome common marketing effectiveness measurement challenges.\u003C\u002Fp>\n\n\n\n\u003Cp>For context, MMM is a cookieless form of measurement which utilises statistical techniques to help you answer three main questions: \u003C\u002Fp>\n\n\n\n\u003Col class=\"wp-block-list\">\n\u003Cli>To what extent are my advertising channels contributing to KPIs?\u003C\u002Fli>\n\n\n\n\u003Cli>What was my return on advertising spend (ROAS)?\u003C\u002Fli>\n\n\n\n\u003Cli>How do I optimise my budget allocation such that my ROAS is maximised?\u003C\u002Fli>\n\u003C\u002Fol>\n\n\n\n\u003Cp>For a comprehensive understanding of MMM and its role in media effectiveness measurement, we recommend diving into our \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fguide-to-media-effectiveness-measurement\u002F\">beginners’ guide to incrementality testing and media mix modelling\u003C\u002Fa>. This guide delves deep into the intricate details of MMM in general, exploring its principles, methodologies, and practical applications.\u003C\u002Fp>\n\n\n\n\u003Cp>Ready to learn more about Google Meridian and what it has to offer? Read on for the expert take compiled by our in-house \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Fmeasurement\u002F\">measurement consultants\u003C\u002Fa>.\u003C\u002Fp>\n\n\n\n\u003Cdiv class=\"table-of-contents\">\u003Cul>\u003Cli class=\"heading-h2\">\u003Ca href=\"#key-out-of-the-box-features-meridian-has-to-offer\">Key out-of-the-box features Meridian has to offer\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#things-to-be-cautious-about-regarding-meridian\">Things to be cautious about regarding Meridian\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#lightweight-mmm-discontinued\">Lightweight MMM discontinued\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#google-meridians-four-pillars\">Google Meridian’s four pillars\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#alternative-mmm-platforms-to-consider\">Alternative MMM platforms to consider\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Ful>\u003C\u002Fdiv>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Fmedia-effectiveness\u002F\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"3840\" height=\"540\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002FMeasurement_Banner_01.jpg?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-27690\"\u002F>\u003C\u002Fa>\u003C\u002Ffigure>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"key-out-of-the-box-features-meridian-has-to-offer\">\u003Cstrong>Key out-of-the-box features Meridian has to offer\u003C\u002Fstrong>\u003C\u002Fh2>\n\n\n\n\u003Cp>\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fdevelopers.google.com\u002Fmeridian\u002Fdocs\u002Fbasics\u002Fbayesian-inference\">\u003Cstrong>Utilises Bayesian Regression:\u003C\u002Fstrong>\u003C\u002Fa> By using Bayesian methods, Meridian can quantify any uncertainty associated with advertising contributions more effectively and make informed decisions based on these intervals of uncertainty. This approach allows for flexible modelling in cases where data is limited and the complexities of your business make the incorporation of prior knowledge crucial. Ultimately Bayesian modelling can provide richer insights into the underlying processes.\u003C\u002Fp>\n\n\n\n\u003Cp>\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fresearch.google\u002Fpubs\u002Fgeo-level-bayesian-hierarchical-media-mix-modeling\u002F\">\u003Cstrong>Geo-based hierarchical modelling:\u003C\u002Fstrong>\u003C\u002Fa> Meridian allows for the incorporation of geo-level advertising data, which likely contains much more information about your marketing effectiveness than national-level data. By using a geo-based hierarchy, geographies can share information with each other. This improves uncertainty when modelling geographies with limited data or very little prior knowledge. Meridian will support models ranging from 50+ geographies to the standard national-level approach if geo-level data isn’t available.\u003C\u002Fp>\n\n\n\n\u003Cp>\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fdevelopers.google.com\u002Fmeridian\u002Fdocs\u002Fbasics\u002Froi-priors\">\u003Cstrong>Uses ROI priors\u003C\u002Fstrong>:\u003C\u002Fa> The Meridian model structure can be re-parameterized such that the ROI of each channel is a model parameter. This enables insights from incrementality tests or industry benchmarks to be used for calibration. Google has announced a slight revamping of their incrementality tests such that user-based and geo-based tests are now going to be referred to as “conversion lift” in Google ads. As these new “conversion lift” tests are going to use user-level data, it’ll be important to ensure that advanced MarTech tracking is in place such as \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fimprove-conversion-accuracy-with-enhanced-conversions\u002F\">Enhanced Conversions\u003C\u002Fa> and \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fserver-side-tagging-gtm\u002F\">Server Side Tagging\u003C\u002Fa>, this will ensure your results are as accurate as possible as cookies are depreciated.\u003C\u002Fp>\n\n\n\n\u003Cp>\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fresearch.google\u002Fpubs\u002Fbayesian-hierarchical-media-mix-model-incorporating-reach-and-frequency-data\u002F\">\u003Cstrong>Incorporates reach and frequency:\u003C\u002Fstrong>\u003C\u002Fa>\u003Cstrong> \u003C\u002Fstrong>Reach refers to the number of users that see an ad within each period. Frequency refers to the corresponding average number of times a given user sees an ad.\u003Cstrong> \u003C\u002Fstrong>Meridian provides the option to use reach and frequency data as model inputs to provide additional insights into the performance of particular media channels. Google claims that currently, MMM is more budget-focused as opposed to outcomes\u002FROI-focused, incorporation of reach and frequency data, along with ROI priors should help to tackle this.\u003C\u002Fp>\n\n\n\n\u003Cp>\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2106.03322\">\u003Cstrong>Time-variant intercept:\u003C\u002Fstrong>\u003C\u002Fa>\u003Cstrong> \u003C\u002Fstrong>The intercept in an MMM describes what your baseline KPIs would be in the absence of any contributing variables such as media activity, non-marketing factors or seasonality. In traditional MMM the intercept is modelled as constant, meaning that the idea that baseline sales can fluctuate over time is neglected. Meridian allows a time-variant intercept, hence allowing for a more flexible way to capture brand awareness or general business growth.\u003C\u002Fp>\n\n\n\n\u003Cp>Dashboard output: Based on our recent visit to Google’s Think Measurement event, we know that once Meridian becomes available for public use we’ll see model outputs in terms of a dashboard. We believe that this will be similar to the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002Fdocs\u002Ffeatures\u002F#model-onepager\">1-pager outputs from Meta’s Robyn\u003C\u002Fa>. \u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"things-to-be-cautious-about-regarding-meridian\">\u003Cstrong>Things to be cautious about regarding Meridian\u003C\u002Fstrong>\u003C\u002Fh2>\n\n\n\n\u003Cp>\u003Cstrong>Does not allow for time-variant media parameters:\u003C\u002Fstrong> Meridian neglects the idea that media contributions can fluctuate over time. Admittedly making every media channel time variant may be unnecessary and cause some serious overfitting problems. However not having the ability to experiment with this or to apply a time-variant contribution to a particular channel is a major drawback.\u003C\u002Fp>\n\n\n\n\u003Cp>\u003Cstrong>Limited prior customisation: \u003C\u002Fstrong>Based on \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fdevelopers.google.com\u002Fmeridian\u002Fdocs\u002Fbasics\u002Fdefault-prior-distributions\">recent documentation\u003C\u002Fa>, it isn’t yet obvious if Meridian will allow a wide range of customisable prior distributions. Capturing the complexities of a specific business using a “one prior fits all” approach might not work for all business types. Meridian has hinted that in the future a wider range of priors might be added, but no timeline was attached to this claim.\u003C\u002Fp>\n\n\n\n\u003Cp>\u003Cstrong>Geo-based hierarchy with limited geos:\u003C\u002Fstrong> Not all businesses will advertise in a large number of countries. Some businesses might only advertise in two or three countries. If these types of businesses were to model using a geo-based hierarchy, then they might run into some sampling problems. This is because hierarchical models typically require a larger number of levels to achieve reliable estimates.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"lightweight-mmm-discontinued\">\u003Cstrong>Lightweight MMM discontinued\u003C\u002Fstrong>\u003C\u002Fh2>\n\n\n\n\u003Cp>Before the release of Meridian, Google’s unofficial software solution for media effectiveness measurement was \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Flightweight-mmm.readthedocs.io\u002Fen\u002Flatest\u002F\">LightWeight MMM\u003C\u002Fa>. Lightweight MMM is an MMM library which also utilises Bayesian statistics and adopts a geography-based hierarchical structure. \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fgithub.com\u002Fgoogle\u002Flightweight_mmm\">Here\u003C\u002Fa>, Google has announced that once Meridian reaches public availability, lightweight MMM will be discontinued.\u003C\u002Fp>\n\n\n\n\u003Cp>The discontinuation of Google’s unofficial product, Lightweight MMM, in light of the forthcoming release of Meridian, heralds a notable transition for users of the former. This shift signals a significant change, as Meridian, leveraging \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.tensorflow.org\u002Fprobability\">TensorFlow Probability\u003C\u002Fa> over \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fnum.pyro.ai\u002Fen\u002Flatest\u002Fgetting_started.html\">Numpyro\u003C\u002Fa> in Lightweight MMM, presents an alternative approach to MMM. This development suggests that users familiar with Lightweight MMM will need to adjust to a new system that diverges considerably from its predecessor in methodology and user experience.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"google-meridians-four-pillars\">\u003Cstrong>Google Meridian’s four pillars\u003C\u002Fstrong>\u003C\u002Fh2>\n\n\n\n\u003Cp>Google Meridan claims to be sold on the following four pillars:\u003C\u002Fp>\n\n\n\n\u003Cp>\u003Cstrong>\u003Cspan style=\"text-decoration: underline;\">Innovation:\u003C\u002Fspan>\u003C\u002Fstrong>\u003C\u002Fp>\n\n\n\n\u003Cul class=\"wp-block-list\">\n\u003Cli>Combines video planning with outcomes – YouTube reach and frequency input data.\u003C\u002Fli>\n\n\n\n\u003Cli>Calibration with experiments – feeds causality into the MMM.\u003C\u002Fli>\n\n\n\n\u003Cli>Measuring search – incorporation of Google search volume (GSV).\u003C\u002Fli>\n\u003C\u002Ful>\n\n\n\n\u003Cp>\u003Cstrong>\u003Cspan style=\"text-decoration: underline;\">Transparency:\u003C\u002Fspan>\u003C\u002Fstrong>\u003C\u002Fp>\n\n\n\n\u003Cul class=\"wp-block-list\">\n\u003Cli>OSS – increased efficiency and reduced manual effort.\u003C\u002Fli>\n\n\n\n\u003Cli>Customisable – recognises that a one-model-fits-all isn’t always the best approach to MMM.\u003C\u002Fli>\n\u003C\u002Ful>\n\n\n\n\u003Cp>\u003Cstrong>\u003Cspan style=\"text-decoration: underline;\">Actionability\u003C\u002Fspan>:\u003C\u002Fstrong>\u003C\u002Fp>\n\n\n\n\u003Cul class=\"wp-block-list\">\n\u003Cli>Enables actionable decisions – optimal frequency such that ROI is maximised.\u003C\u002Fli>\n\n\n\n\u003Cli>Provides the most impactful media mix under constrained optimisation.\u003C\u002Fli>\n\n\n\n\u003Cli>Full technical documentation on academia & how to use and implement\u003C\u002Fli>\n\u003C\u002Ful>\n\n\n\n\u003Cp>\u003Cstrong>\u003Cspan style=\"text-decoration: underline;\">Education:\u003C\u002Fspan>\u003C\u002Fstrong>\u003C\u002Fp>\n\n\n\n\u003Cul class=\"wp-block-list\">\n\u003Cli>Full technical documentation on academia and implementation. For those with early access, more documentation is available.\u003C\u002Fli>\n\u003C\u002Ful>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"alternative-mmm-platforms-to-consider\">\u003Cstrong>Alternative MMM platforms to consider\u003C\u002Fstrong>\u003C\u002Fh2>\n\n\n\n\u003Cp>You can \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fdevelopers.google.com\u002Fmeridian\">apply for early access to Google Meridian\u003C\u002Fa> now, but whilst you wait there are a handful of different open-source MMM platforms out there which are ready for use right now. The most popular include Meta’s \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Ffacebookexperimental.github.io\u002FRobyn\u002Fdocs\u002Fwelcome\">Robyn\u003C\u002Fa>, \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.pymc-labs.com\u002F\">PyMC\u003C\u002Fa> and \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.pymc-marketing.io\u002Fen\u002Fstable\u002F\">PyMC marketing\u003C\u002Fa>.\u003C\u002Fp>\n\n\n\n\u003Cp>See some of the main comparisons tabulated below:\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-table\">\u003Ctable>\u003Ctbody>\u003Ctr>\u003Ctd>\u003C\u002Ftd>\u003Ctd>\u003Cstrong>Google’s Meridian\u003C\u002Fstrong>\u003C\u002Ftd>\u003Ctd>\u003Cstrong>PyMC – Marketing\u003C\u002Fstrong>\u003C\u002Ftd>\u003Ctd>\u003Cstrong>PyMC – Custom Modelling\u003C\u002Fstrong>\u003C\u002Ftd>\u003Ctd>\u003Cstrong>Meta’s Robyn\u003C\u002Fstrong>\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Is it Bayesian?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Allows for geo-hierarchical modelling?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Allows for channel-hierarchical modelling?\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Allows for calibration?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Allows for a Time-variant intercept?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Allows Time-variant media channels?\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Incorporates reach and frequency?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Flexible prior customisation?\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>\u003Cstrong>N\u002FA\u003C\u002Fstrong>\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Accounts for delayed effects of advertising?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Accounts for diminishing returns effects?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Out of sample prediction?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Constrained Budget Allocation?\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Easy to implement?\u003C\u002Ftd>\u003Ctd>\u003Cstrong>TBD\u003C\u002Fstrong>\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>Documentation?\u003C\u002Ftd>\u003Ctd>❌\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003Ctd>✅\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Ffigure>\n\n\n\n\u003Cp>At Impression, we advocate for a tailored approach to measurement, understanding that a \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.warc.com\u002Fnewsandopinion\u002Fopinion\u002Fan-open-letter-to-marketing-departments-from-experts-on-evaluating-advertising\u002Fen-gb\u002F6261\">one-size-fits-all model\u003C\u002Fa> may not capture the intricacies of diverse marketing strategies. That’s why we’ve developed our bespoke custom media mix model using cutting-edge tools from PyMC labs, ensuring adaptability to the dynamic digital marketing landscape while delivering precise insights tailored to each client’s unique needs. \u003C\u002Fp>\n\n\n\n\u003Cp>As you can probably tell, our preference is PyMC as we feel it allows us to construct, implement and analyse complex probabilistic models with ease. Although it requires some statistical and computational knowledge to fully understand how to implement MMM in PyMC, the extensive documentation and notebooks that PyMC has to offer make it accessible to both beginners and seasoned practitioners in the field. With its robust capabilities for \u003Ca rel=\"noreferrer noopener\" href=\"http:\u002F\u002Fgithub.com\u002Fpymc-labs\u002Fpymc-marketing\u002Fpull\u002F482(opens in a new tab)\">time-variant par\u003C\u002Fa>\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fgithub.com\u002Fpymc-labs\u002Fpymc-marketing\u002Fpull\u002F482\">ameters\u003C\u002Fa>, \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.pymc-labs.com\u002Fblog-posts\u002Ffrom-uncertainty-to-insight-how-bayesian-data-science-can-transform-your-business\u002F\">uncertainty quantification\u003C\u002Fa>, and \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.pymc.io\u002Fprojects\u002Fexamples\u002Fen\u002Flatest\u002Fgeneralized_linear_models\u002FGLM-out-of-sample-predictions.html\">out-of-sample prediction\u003C\u002Fa>, we feel PyMC is the optimal available tool that facilitates sophisticated analysis and decision-making in the digital marketing domain.\u003C\u002Fp>\n\n\n\n\u003Cp>As we eagerly await the broader release of Google’s Meridian, the true practicality of it remains uncertain. While documentation hints at statistically robust foundations of a similar nature to other established MMM libraries, the full extent of its capabilities and ease of integration will only become clear with wider adoption and access to its source code.\u003C\u002Fp>\n\n\n\n\u003Cp>If you are looking for support with conducting media mix modelling for your business, don’t hesitate to \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fget-in-touch\u002F\">get in touch\u003C\u002Fa> with our data science team.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Ca rel=\"noreferrer noopener\" href=\"\u002Fget-in-touch\u002F\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"3840\" height=\"540\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002FGeneric_Banner_01.jpg?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-27527\"\u002F>\u003C\u002Fa>\u003C\u002Ffigure>\n",false,{"rendered":115,"protected":211},29,27608,"standard",{"_yoast_wpseo_focuskw":5,"_yoast_wpseo_title":114,"_yoast_wpseo_metadesc":217,"inline_featured_image":211,"footnotes":5},"Discover Google Meridian, an open-source Media Mix Modelling (MMM) platform that deepens your understanding of marketing effectiveness.",[219,220],39,41,[],[223],120,[225,113,226,227,228,229,230,231,232],"post-27600","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-digital-marketing","category-measurement",[],[],[236,237,240,241],{"path":185,"name":186},{"path":238,"name":239},"\u002Fblog\u002F","Blog",{"path":124,"name":125},{"path":194,"name":114},{"thumbnail":243,"medium":245,"medium_large":247,"large":249,"1536x1536":253,"2048x2048":255,"post-thumbnail":257,"12_col":259,"default_landscape":261,"default_portrait":263,"6_col_and_margin_short":265,"11_col_and_margin":267,"9_col_rectangle":269,"11_col":271,"staff_image_696_600":273,"rectangular_thumbnail":275},{"url":119,"width":244,"height":244},640,{"url":246,"width":70,"height":70},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=113&q=90&w=200",{"url":248,"width":73,"height":23},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=432&q=90&w=768",{"url":250,"width":251,"height":252},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1000&q=90&w=1892",1892,1000,{"url":254,"width":79,"height":79},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=864&q=90&w=1536",{"url":256,"width":82,"height":82},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1152&q=90&w=2048",{"url":258,"width":85,"height":86},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1180&q=90&w=1860",{"url":260,"width":88,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2880",{"url":262,"width":91,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=800",{"url":264,"width":65,"height":91},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=800&q=90&w=600",{"url":266,"width":85,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=1860",{"url":268,"width":95,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=3080",{"url":270,"width":97,"height":98},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1400&q=90&w=2150",{"url":272,"width":100,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2640",{"url":274,"width":64,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=696",{"url":276,"width":103,"height":104},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=300&q=90&w=750",[278],{"id":213,"path":279,"display_name":4,"job_title":11,"description":6,"avatar_url":122,"acf":280},"\u002Fabout\u002Four-team\u002Fharry-brace\u002F",{"full_bio":10,"job_title":11,"specialist_service_links":281,"photo":288},[282,284,286],{"page":283,"link_text":29},{"ID":15,"post_author":16,"post_date":17,"post_date_gmt":17,"post_content":5,"post_title":18,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":21,"to_ping":5,"pinged":5,"post_modified":22,"post_modified_gmt":22,"post_content_filtered":5,"post_parent":23,"guid":24,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":28},{"page":285,"link_text":34},{"ID":32,"post_author":16,"post_date":33,"post_date_gmt":33,"post_content":5,"post_title":34,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":35,"to_ping":5,"pinged":5,"post_modified":36,"post_modified_gmt":36,"post_content_filtered":5,"post_parent":37,"guid":38,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":39},{"page":287,"link_text":44},{"ID":42,"post_author":16,"post_date":43,"post_date_gmt":43,"post_content":5,"post_title":44,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":45,"to_ping":5,"pinged":5,"post_modified":46,"post_modified_gmt":46,"post_content_filtered":5,"post_parent":37,"guid":47,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":48},{"ID":50,"id":50,"title":51,"filename":52,"filesize":53,"url":54,"link":55,"alt":5,"author":56,"description":5,"caption":5,"name":57,"status":58,"uploaded_to":23,"date":59,"modified":59,"menu_order":23,"mime_type":60,"type":61,"subtype":62,"icon":63,"width":64,"height":65,"sizes":289},{"thumbnail":67,"thumbnail-width":68,"thumbnail-height":68,"medium":69,"medium-width":70,"medium-height":71,"medium_large":72,"medium_large-width":73,"medium_large-height":74,"large":75,"large-width":76,"large-height":77,"1536x1536":78,"1536x1536-width":79,"1536x1536-height":80,"2048x2048":81,"2048x2048-width":82,"2048x2048-height":83,"post-thumbnail":84,"post-thumbnail-width":85,"post-thumbnail-height":86,"12_col":87,"12_col-width":88,"12_col-height":89,"default_landscape":90,"default_landscape-width":91,"default_landscape-height":65,"default_portrait":92,"default_portrait-width":65,"default_portrait-height":91,"6_col_and_margin_short":93,"6_col_and_margin_short-width":85,"6_col_and_margin_short-height":89,"11_col_and_margin":94,"11_col_and_margin-width":95,"11_col_and_margin-height":89,"9_col_rectangle":96,"9_col_rectangle-width":97,"9_col_rectangle-height":98,"11_col":99,"11_col-width":100,"11_col-height":89,"staff_image_696_600":101,"staff_image_696_600-width":64,"staff_image_696_600-height":65,"rectangular_thumbnail":102,"rectangular_thumbnail-width":103,"rectangular_thumbnail-height":104},"Article",[292,293],{"id":219,"name":125,"path":124},{"id":220,"name":136,"path":135},[295,297,299,302,304,307,310,312,314,316,319,322,325,328,331,334,337,339,340,343,345,348],{"property":296,"content":114},"title",{"property":298,"content":217},"description",{"name":300,"content":301},"robots","index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1",{"property":199,"content":303},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgoogle-meridian-what-you-need-to-know\u002F",{"name":305,"content":306},"og:locale","en_GB",{"property":308,"content":309},"og:type","article",{"property":311,"content":114},"og:title",{"property":313,"content":217},"og:description",{"property":315,"content":303},"og:url",{"property":317,"content":318},"og:site_name","Impression",{"property":320,"content":321},"article:publisher","https:\u002F\u002Fwww.facebook.com\u002Fimpression",{"property":323,"content":324},"article:published_time","2025-01-17T09:00:18+00:00",{"property":326,"content":327},"article:modified_time","2025-01-17T09:13:55+00:00",{"property":329,"content":330},"og:image","https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fopen-graph-google-meridian.png?auto=compress%2Cformat&q=90",{"property":332,"content":333},"og:image:width",1200,{"property":335,"content":336},"og:image:height",630,{"property":338,"content":60},"og:image:type",{"property":180,"content":4},{"property":341,"content":342},"twitter:card","summary_large_image",{"property":344,"content":330},"twitter:image",{"property":346,"content":347},"twitter:creator","@impressiontalk",{"property":349,"content":347},"twitter:site",{"@context":351,"@graph":352},"https:\u002F\u002Fschema.org",[353,367,376,380,395,405,471],{"@type":290,"@id":354,"isPartOf":355,"author":356,"headline":114,"datePublished":324,"dateModified":327,"mainEntityOfPage":358,"wordCount":359,"publisher":360,"image":362,"thumbnailUrl":364,"articleSection":365,"inLanguage":366},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgoogle-meridian-what-you-need-to-know\u002F#article",{"@id":303},{"name":4,"@id":357},"https:\u002F\u002Fadmin.impressiondigital.com\u002F#\u002Fschema\u002Fperson\u002Fa18b79c97abbcb2444ace02aa88250ae",{"@id":303},1434,{"@id":361},"https:\u002F\u002Fadmin.impressiondigital.com\u002F#organization",{"@id":363},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgoogle-meridian-what-you-need-to-know\u002F#primaryimage","https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002Fbastian-riccardi-0uMU42uRw28-unsplash.jpg?auto=compress%2Cformat&q=90",[125,136],"en-GB",{"@type":368,"@id":303,"url":303,"name":114,"isPartOf":369,"primaryImageOfPage":371,"image":372,"thumbnailUrl":364,"datePublished":324,"dateModified":327,"description":217,"breadcrumb":373,"inLanguage":375},"WebPage",{"@id":370},"https:\u002F\u002Fwww.impressiondigital.com\u002F#website",{"@id":363},{"@id":363},{"@id":374},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgoogle-meridian-what-you-need-to-know\u002F#breadcrumb","en",{"@type":377,"inLanguage":366,"@id":363,"url":364,"contentUrl":364,"width":378,"height":379},"ImageObject",2400,1350,{"@type":381,"@id":374,"itemListElement":382},"BreadcrumbList",[383,387,390,393],{"@type":384,"position":385,"name":186,"item":386},"ListItem",1,"https:\u002F\u002Fwww.impressiondigital.com\u002F",{"@type":384,"position":388,"name":239,"item":389},2,"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002F",{"@type":384,"position":391,"name":125,"item":392},3,"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fcategory\u002Fdigital-marketing\u002F",{"@type":384,"position":394,"name":114},4,{"@type":396,"@id":370,"url":386,"name":318,"description":5,"publisher":397,"inLanguage":366,"potentialAction":399},"WebSite",{"@id":398},"https:\u002F\u002Fwww.impressiondigital.com\u002F#organization",{"@type":400,"target":401,"query-input":404},"SearchAction",{"@type":402,"urlTemplate":403},"EntryPoint","https:\u002F\u002Fwww.impressiondigital.com\u002Fsearch?q={search_term_string}","required name=search_term_string",{"@type":406,"@id":398,"name":318,"url":386,"logo":407,"image":412,"sameAs":413,"@context":417,"email":418,"foundingDate":419,"legalName":420,"vatID":421,"description":422,"telephone":423,"address":424,"location":431,"areaServed":464},"Organization",{"@type":377,"inLanguage":366,"@id":408,"url":409,"contentUrl":409,"width":410,"height":411,"caption":318},"https:\u002F\u002Fwww.impressiondigital.com\u002F#\u002Fschema\u002Flogo\u002Fimage\u002F","https:\u002F\u002Fcdn.impression.co.uk\u002F2021\u002F06\u002FA5LuTDrq-Impression_Wordmark.png",920,176,"https:\u002F\u002Fcdn.impression.co.uk\u002F2021\u002F04\u002FFAVxDeOS-Nottingham.jpg",[414,415,416],"https:\u002F\u002Fwww.linkedin.com\u002Fcompany\u002Fimpression-digital-limited","https:\u002F\u002Fwww.facebook.com\u002Fimpression\u002F","https:\u002F\u002Fwww.instagram.com\u002Fimpressiontalk\u002F","https:\u002F\u002Fwww.schema.org\u002F","hello@impressiondigital.com","2012-11-07","Impression Digital Limited","GB186363774","Impression is an award-winning digital marketing agency based in the UK and USA. We offer SEO agency, PPC agency, Digital PR agency and Analytics services to companies of all sizes across the UK, North America, Europe and beyond. Get in touch with us if you think we can work together.","01158 242 212",{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},"PostalAddress","Fothergill House, 16 King Street","Nottingham","Nottinghamshire","NG1 2AS","United Kingdom",[432,441,453],{"@type":433,"name":434,"geo":435,"address":439,"photo":412,"telephone":440},"Place","Impression Digital - Nottingham",{"@type":436,"latitude":437,"longitude":438},"GeoCoordinates","52.95501401025914","-1.1494887460285232",{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},"+44 01158 242 212",{"@type":433,"name":442,"geo":443,"address":446,"photo":451,"telephone":452},"Impression Digital - London",{"@type":436,"latitude":444,"longitude":445},"51.50237208377361","-0.09229841534284115",{"@type":425,"streetAddress":447,"addressLocality":448,"addressRegion":449,"postalCode":450,"addressCountry":430},"201 Borough High St","Southwark","London","SE1 1JA","https:\u002F\u002Fcdn.impression.co.uk\u002F2021\u002F04\u002FwD0pPVWz-London.jpg","+44 020 7112 8148",{"@type":433,"name":454,"geo":455,"address":458},"Impression Digital - New York",{"@type":436,"latitude":456,"longitude":457},"40.726722746949","-74.00527778220652",{"@type":425,"streetAddress":459,"addressLocality":460,"addressRegion":461,"postalCode":462,"addressCountry":463},"160 Varick St","New York","NY","10013","USA",[465,468,470],{"@type":466,"name":467},"Country","Worldwide",{"@type":466,"name":469},"GBR",{"@type":466,"name":463},{"@type":472,"@id":473,"name":4,"image":474,"description":6,"sameAs":476,"url":200},"Person","https:\u002F\u002Fwww.impressiondigital.com\u002F#\u002Fschema\u002Fperson\u002Fa18b79c97abbcb2444ace02aa88250ae",{"@type":377,"inLanguage":366,"@id":475,"url":8,"contentUrl":8,"caption":4},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress&fit=fit&fm=jpg&h=96&q=90&w=9615847ead018a2de28b7216e39fc92f32",[196],{"self":478,"collection":488,"about":491,"author":494,"replies":497,"version-history":500,"predecessor-version":504,"wp:featuredmedia":508,"wp:attachment":511,"wp:term":514,"curies":523},[479],{"href":480,"targetHints":481},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F27600",{"allow":482},[483,484,485,486,487],"GET","POST","PUT","PATCH","DELETE",[489],{"href":490},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts",[492],{"href":493},"\u002Fwp-json\u002Fwp\u002Fv2\u002Ftypes\u002Fpost",[495],{"embeddable":108,"href":496},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fusers\u002F29",[498],{"embeddable":108,"href":499},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcomments?post=27600",[501],{"count":502,"href":503},17,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F27600\u002Frevisions",[505],{"id":506,"href":507},31543,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F27600\u002Frevisions\u002F31543",[509],{"embeddable":108,"href":510},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia\u002F27608",[512],{"href":513},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia?parent=27600",[515,518,521],{"taxonomy":516,"embeddable":108,"href":517},"category","\u002Fwp-json\u002Fwp\u002Fv2\u002Fcategories?post=27600",{"taxonomy":519,"embeddable":108,"href":520},"collections","\u002Fwp-json\u002Fwp\u002Fv2\u002Fcollections?post=27600",{"taxonomy":180,"embeddable":108,"href":522},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcoauthors?post=27600",[524],{"name":525,"href":526,"templated":108},"wp","https:\u002F\u002Fapi.w.org\u002F{rel}",{"id":528,"date":529,"date_gmt":529,"guid":530,"modified":130,"modified_gmt":130,"slug":532,"status":19,"type":113,"link":129,"title":533,"content":534,"excerpt":536,"author":213,"featured_media":537,"comment_status":20,"ping_status":20,"sticky":211,"template":5,"format":215,"meta":538,"categories":540,"collections":541,"coauthors":542,"class_list":543,"acf":545,"head_links":546,"breadcrumbs":547,"url_path":129,"search_description":539,"featured_image_urls":552,"authors":584,"type_nicename":290,"reading_minutes":118,"category":596,"yoast_title":127,"yoast_meta":598,"yoast_json_ld":625,"_links":678},25590,"2025-01-08T09:43:50",{"rendered":531},"\u002F?p=25590","server-side-tagging-gtm",{"rendered":127},{"rendered":535,"protected":211},"\n\u003Cp>Server-side tagging is now available within Google Tag Manager (GTM), putting better marketing effectiveness data back into your hands. Packed with tips from our in-house experts, this guide unpacks the principles of server-side tagging and how you can use it to evolve your measurement strategy in GTM. \u003C\u002Fp>\n\n\n\n\u003Cp>Read on to discover the benefits of this approach, and ways to overcome common limitations that our \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fanalytics\u002F\">analytics consultants\u003C\u002Fa> have encountered. Need support with implementing server-side tagging in GTM? \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fget-in-touch\u002F\">Contact our team today\u003C\u002Fa> for a free consultation.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"what-is-server-side-tagging-and-why-consider-it\">\u003Cstrong>What is server-side tagging, and why consider it? \u003C\u002Fstrong>\u003C\u002Fh2>\n\n\n\n\u003Cp>In recent years, the ability for web and ad analytics platforms to capture the page and attribution data we are used to has been decreasing. There are a number of reasons for this, including technical changes driven by privacy regulations, and in general, the response has been for tech vendors to collect less personally identifiable information. \u003C\u002Fp>\n\n\n\n\u003Cp>At the same time, these technological changes have begun to have lasting impacts on the efficacy of the collection mechanisms previously used. \u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image\">\u003Cimg decoding=\"async\" src=\"https:\u002F\u002Flh7-eu.googleusercontent.com\u002Fdocsz\u002FAD_4nXd5Znkv8LUpr9PVBwsX8CYxWIE3CzjZRzc1keApCrcJcohQHE_yQQh9Cb-CohO2WdLga1sSDX-qxiJWHKV0nbe3czo9C32Lb3XEzpjOZRiYkZDLf8uv0lX88iSDwJz-JFoHyrKRz6fyHeKRALnCIM0l-Daq?key=9-A8IvgWzV-lJc6gYn5F1w\" alt=\"\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Cp>Collecting your ad tech and analytics through a server-side solution, such as Server-Side Google Tag Manager can give you many benefits, including: \u003C\u002Fp>\n\n\n\n\u003Col class=\"wp-block-list\">\n\u003Cli>Improve data collection, including conversion tracking\n\u003Col class=\"wp-block-list\">\n\u003Cli>Reduce the impact of ad blockers by moving to a first-party context \u003C\u002Fli>\n\n\n\n\u003Cli>Extend cookie lifetime beyond third-party cookie browser limitations\u003C\u002Fli>\n\u003C\u002Fol>\n\u003C\u002Fli>\n\n\n\n\u003Cli>Better data governance and control\n\u003Col class=\"wp-block-list\">\n\u003Cli>Control which data is sent to each ad tech and martech vendor \u003C\u002Fli>\n\n\n\n\u003Cli>Enrich your data (via APIs and your first-party databases), but do so in a private and safe manner \u003C\u002Fli>\n\u003C\u002Fol>\n\u003C\u002Fli>\n\n\n\n\u003Cli>Reduced browser load; Increase page speed\n\u003Col class=\"wp-block-list\">\n\u003Cli>Send data once to your container, then from there to multiple platforms \u003C\u002Fli>\n\n\n\n\u003Cli>Moving tagging server-side can improve page load speeds by up to 7% (\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fmarketingplatform.google.com\u002Fabout\u002Fresources\u002Fnemlig-gathers-new-insights-and-grows-conversions\u002F\">Case study\u003C\u002Fa>)\u003C\u002Fli>\n\u003C\u002Fol>\n\u003C\u002Fli>\n\u003C\u002Fol>\n\n\n\n\u003Cp>If these benefits sound like they may help solve some of the challenges you have with your marketing data collection and accuracy, please \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fget-in-touch\u002F\">get in touch\u003C\u002Fa> with our Media Solutions team.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"how-does-server-side-tagging-work\">How does server-side tagging work?\u003C\u002Fh2>\n\n\n\n\u003Cp>Before running through the benefits, let’s take a look at how server-side tagging works and how it is different from traditional Google Tag Manager tracking.\u003C\u002Fp>\n\n\n\n\u003Cp>Traditionally, Google Tag Manager has been 100% client-side. This means tracking scripts will run in the user’s browser when required and any related javascript files will be downloaded. Often these scripts will then set third-party cookies and begin sending data to the related marketing platforms directly from the browser.  \u003C\u002Fp>\n\n\n\n\u003Cp>Visualised below is a typical example of a website without Server-Side tagging. It shows how each event is sent directly to each platform directly from the browser.\u003C\u002Fp>\n\n\n\u003Cdiv class=\"wp-block-image\">\n\u003Cfigure class=\"aligncenter size-full is-resized\">\u003Cimg loading=\"lazy\" decoding=\"async\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fserver-side-tag-management-GTM-1.jpg?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-25618\" width=\"610\" height=\"610\"\u002F>\u003C\u002Ffigure>\u003C\u002Fdiv>\n\n\n\u003Cp>With Server-Side tagging, this flow of data is different. A standard GTM container is still required, however this time, it works in parallel with an intermediary server-side container hosted with a provider of your choice. Instead of downloading and running multiple tracking scripts from your standard GTM container (often carrying out very similar jobs), you now only need to send one signal to your Server Side container. This can then be distributed directly to any platform that needs the data. This could be anything from a page view to a purchase event.\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">\u003Cem>Website configuration with server-side Tag Manager:\u003C\u002Fem>\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1999\" height=\"1143\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fserver-side-tag-management-GTM-2.jpg?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-25619\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"benefits-of-server-side-tagging-in-google-tag-manager\">Benefits of server-side tagging in Google Tag Manager\u003C\u002Fh2>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"simplified-implementation-of-conversion-apis\">\u003Cstrong>Simplified implementation of conversion APIs\u003C\u002Fstrong>\u003C\u002Fh3>\n\n\n\n\u003Cp>Utilising a Google Server-Side container makes the implementation of server-side conversion APIs much easier. Currently, out of the box, you can utilise Facebook and Google Ads conversion APIs to send data directly to Google and Facebook Ad servers. This improves the accuracy of your conversion tracking in instances where the user may have an adblocker or is using a browser with enhanced tracking protection such as Safari or Firefox.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"full-control-over-your-data\">\u003Cstrong>Full control over your data\u003C\u002Fstrong>\u003C\u002Fh3>\n\n\n\n\u003Cp>When you place third-party tracking scripts directly on your website, vendors have the power to execute any code and collect any data they see fit. Whilst these are almost always trusted vendors, you still have less control over your data. With server-side tracking, you control exactly what data is collected without any third-party JS libraries being downloaded.\u003C\u002Fp>\n\n\n\n\u003Cp>This level of control is \u003Cstrong>\u003Cem>crucial\u003C\u002Fem>\u003C\u002Fstrong> for safeguarding user privacy and compliance with GDPR, as you can implement strict PII filtering to ensure sensitive user information is not exposed.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"reduced-load-on-customers-browser\">\u003Cstrong>Reduced load on customer’s browser\u003C\u002Fstrong>\u003C\u002Fh3>\n\n\n\n\u003Cp>By moving tracking scripts server side, there is less to be downloaded and executed on the client’s web browser, reducing page loading times and making your website faster.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"first-party-cookie-context\">\u003Cstrong>First-party cookie context\u003C\u002Fstrong>\u003C\u002Fh3>\n\n\n\n\u003Cp>After setting up a server-side \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Ftag-management\u002F\">Google Tag Manager\u003C\u002Fa>, you receive an automatically generated endpoint to send your measurement hits to. This can be changed to a domain of your choice such as a dedicated measurement subdomain e.g. \u003Cem>measurement.impression.co.uk.\u003C\u002Fem>\u003C\u002Fp>\n\n\n\n\u003Cp>This means that any cookies set by your Server-Side GTM container will be first-party, bypassing the restrictions caused by browsers like Safari and Firefox which block them by default. \u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"hide-tracking-ids-and-other-identifiers\">\u003Cstrong>Hide tracking IDs and other identifiers \u003C\u002Fstrong>\u003C\u002Fh3>\n\n\n\n\u003Cp>With client-side tracking, your account IDs are usually exposed and allow spammers to send bad data to your accounts. Whilst this isn’t a common issue, the added protection will ensure your data isn’t contaminated.\u003C\u002Fp>\n\n\n\n\u003Cp>Currently, all these benefits are limited to a set of platforms that support Server-Side tagging. These are \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fanalytics\u002F\">Google Analytics\u003C\u002Fa> (Both UA and GA4), Google Ads and Facebook. Whilst custom scripts can be used for other platforms, these are the officially supported ones. This list will certainly expand over the coming months and years. \u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"limitations-of-server-side-tagging-in-google-tag-manager\">Limitations of server-side tagging in Google Tag Manager\u003C\u002Fh2>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"expertise-and-hosting-costs\">\u003Cstrong>Expertise and Hosting Costs\u003C\u002Fstrong>\u003C\u002Fh3>\n\n\n\n\u003Cp>While Server-Side tagging fits neatly alongside client-side tagging, setting it up requires advanced tag management knowledge and some understanding of server architecture. If you are looking to utilise \u003Cstrong>Google Cloud services\u003C\u002Fstrong> such as Cloud Run, you will be guided through using the setup using a wizard which makes things easier to a degree. For other deployments such as App Engine or manual Docker deployments, you’ll also need experience with them specifically.\u003C\u002Fp>\n\n\n\n\u003Cp>Implementing Server-Side tagging through \u003Cstrong>Cloud Run\u003C\u002Fstrong> is now the default deployment option within Google Tag Manager, replacing App Engine. As previously mentioned, GTM provides an installation wizard that walks you through the process and this has also been switched to Cloud Run. It offers better scalability, and cost-efficiency and makes global deployments of sGTM easier. This makes it an attractive choice for large businesses that have customers visiting their websites from across the globe. Although it’s been replaced, \u003Cstrong>App Engine\u003C\u002Fstrong> is cost-free for testing purposes which does provide a convenient starting point for beginners and users wanting to have more flexibility with testing. You then still have the option to upgrade to a production setup as needed.\u003C\u002Fp>\n\n\n\n\u003Cp>An alternative deployment option is to use a \u003Cstrong>Docker\u003C\u002Fstrong> container which offers a high degree of control over your server-side tagging infrastructure. While it provides full autonomy over your setup, the process can become intricate due to the requirements for you to manage your auto-scaling infrastructure – however, this may be preferable for some enterprise organisations that have strict data processing rules. It’s important to note that this approach can incur costs depending on the chosen service infrastructure and may require careful management to optimise expenses.\u003C\u002Fp>\n\n\n\n\u003Cp>Last but not least you have the option to use third-party providers who have their server architecture specifically designed for server-side GTM (sGTM). Providers such as Owntag.eu and Stape.io offer bespoke hosting packages just for sGTM. They also have clear pricing structures based on the number of events you need your sGTM to process each month.\u003C\u002Fp>\n\n\n\n\u003Cp>These services are completely web-based and therefore incredibly easy to set up. After signing up for a specific package, you are then provided with DNS settings and instructions that guide you through all the steps required to get your sGTM setup ready to receive events.\u003C\u002Fp>\n\n\n\n\u003Cp>Another key benefit of using these third-party services is that they offer additional features such as cookie extenders and custom loaders that will further improve your data quality. These features you don’t get with other hosting options and they would be challenging to implement manually. \u003C\u002Fp>\n\n\n\n\u003Cp>Ultimately, whatever hosting option you choose depends on your technical expertise, budget and data processing rules your business needs to adhere to\u003C\u002Fp>\n\n\n\n\u003Chr class=\"wp-block-separator has-alpha-channel-opacity is-style-wide\"\u002F>\n\n\n\n\u003Cp>For more insights and guides on measurement and GA4, be sure to take a look at our \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fcategory\u002Fmeasurement\u002F\">measurement blog\u003C\u002Fa>. If you want to talk about your business aspirations, or explore how our \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fanalytics\u002Fgoogle-tag-manager\u002F\">Google Tag Manager services\u003C\u002Fa> can help, \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fget-in-touch\u002F\">get in touch!\u003C\u002Fa>\u003Ca rel=\"noreferrer noopener\" href=\"javascript:void(0)\">\u003C\u002Fa>\u003Ca rel=\"noreferrer noopener\" href=\"javascript:void(0)\">\u003C\u002Fa>\u003Ca rel=\"noreferrer noopener\" href=\"javascript:void(0)\">\u003C\u002Fa>\u003C\u002Fp>\n",{"rendered":128,"protected":211},25622,{"_yoast_wpseo_focuskw":5,"_yoast_wpseo_title":127,"_yoast_wpseo_metadesc":539,"inline_featured_image":211,"footnotes":5},"Server-side tagging has recently been made available within Google Tag Manager and it comes with various benefits that better put data back into your hands.",[220],[],[223],[544,113,226,227,228,229,230,232],"post-25590",[],[],[548,549,550,551],{"path":185,"name":186},{"path":238,"name":239},{"path":135,"name":136},{"path":194,"name":127},{"thumbnail":553,"medium":554,"medium_large":556,"large":558,"1536x1536":560,"2048x2048":562,"post-thumbnail":564,"12_col":566,"default_landscape":568,"default_portrait":570,"6_col_and_margin_short":572,"11_col_and_margin":574,"9_col_rectangle":576,"11_col":578,"staff_image_696_600":580,"rectangular_thumbnail":582},{"url":131,"width":244,"height":244},{"url":555,"width":70,"height":70},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=152&q=90&w=200",{"url":557,"width":73,"height":23},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=583&q=90&w=768",{"url":559,"width":251,"height":252},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1000&q=90&w=1892",{"url":561,"width":79,"height":79},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1167&q=90&w=1536",{"url":563,"width":82,"height":82},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1556&q=90&w=2048",{"url":565,"width":85,"height":86},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1180&q=90&w=1860",{"url":567,"width":88,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2880",{"url":569,"width":91,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=800",{"url":571,"width":65,"height":91},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=800&q=90&w=600",{"url":573,"width":85,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=1860",{"url":575,"width":95,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=3080",{"url":577,"width":97,"height":98},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1400&q=90&w=2150",{"url":579,"width":100,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2640",{"url":581,"width":64,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=696",{"url":583,"width":103,"height":104},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=300&q=90&w=750",[585],{"id":213,"path":279,"display_name":4,"job_title":11,"description":6,"avatar_url":122,"acf":586},{"full_bio":10,"job_title":11,"specialist_service_links":587,"photo":594},[588,590,592],{"page":589,"link_text":29},{"ID":15,"post_author":16,"post_date":17,"post_date_gmt":17,"post_content":5,"post_title":18,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":21,"to_ping":5,"pinged":5,"post_modified":22,"post_modified_gmt":22,"post_content_filtered":5,"post_parent":23,"guid":24,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":28},{"page":591,"link_text":34},{"ID":32,"post_author":16,"post_date":33,"post_date_gmt":33,"post_content":5,"post_title":34,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":35,"to_ping":5,"pinged":5,"post_modified":36,"post_modified_gmt":36,"post_content_filtered":5,"post_parent":37,"guid":38,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":39},{"page":593,"link_text":44},{"ID":42,"post_author":16,"post_date":43,"post_date_gmt":43,"post_content":5,"post_title":44,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":45,"to_ping":5,"pinged":5,"post_modified":46,"post_modified_gmt":46,"post_content_filtered":5,"post_parent":37,"guid":47,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":48},{"ID":50,"id":50,"title":51,"filename":52,"filesize":53,"url":54,"link":55,"alt":5,"author":56,"description":5,"caption":5,"name":57,"status":58,"uploaded_to":23,"date":59,"modified":59,"menu_order":23,"mime_type":60,"type":61,"subtype":62,"icon":63,"width":64,"height":65,"sizes":595},{"thumbnail":67,"thumbnail-width":68,"thumbnail-height":68,"medium":69,"medium-width":70,"medium-height":71,"medium_large":72,"medium_large-width":73,"medium_large-height":74,"large":75,"large-width":76,"large-height":77,"1536x1536":78,"1536x1536-width":79,"1536x1536-height":80,"2048x2048":81,"2048x2048-width":82,"2048x2048-height":83,"post-thumbnail":84,"post-thumbnail-width":85,"post-thumbnail-height":86,"12_col":87,"12_col-width":88,"12_col-height":89,"default_landscape":90,"default_landscape-width":91,"default_landscape-height":65,"default_portrait":92,"default_portrait-width":65,"default_portrait-height":91,"6_col_and_margin_short":93,"6_col_and_margin_short-width":85,"6_col_and_margin_short-height":89,"11_col_and_margin":94,"11_col_and_margin-width":95,"11_col_and_margin-height":89,"9_col_rectangle":96,"9_col_rectangle-width":97,"9_col_rectangle-height":98,"11_col":99,"11_col-width":100,"11_col-height":89,"staff_image_696_600":101,"staff_image_696_600-width":64,"staff_image_696_600-height":65,"rectangular_thumbnail":102,"rectangular_thumbnail-width":103,"rectangular_thumbnail-height":104},[597],{"id":220,"name":136,"path":135},[599,600,601,602,604,605,606,607,608,609,610,611,613,615,617,618,619,620,621,622,623,624],{"property":296,"content":127},{"property":298,"content":539},{"name":300,"content":301},{"property":199,"content":603},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fserver-side-tagging-gtm\u002F",{"name":305,"content":306},{"property":308,"content":309},{"property":311,"content":127},{"property":313,"content":539},{"property":315,"content":603},{"property":317,"content":318},{"property":320,"content":321},{"property":323,"content":612},"2025-01-08T09:43:50+00:00",{"property":326,"content":614},"2025-03-31T14:07:58+00:00",{"property":329,"content":616},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fopen-graph-server-side-tagging-gtm.png?auto=compress%2Cformat&q=90",{"property":332,"content":333},{"property":335,"content":336},{"property":338,"content":60},{"property":180,"content":4},{"property":341,"content":342},{"property":344,"content":616},{"property":346,"content":347},{"property":349,"content":347},{"@context":351,"@graph":626},[627,638,644,646,653,657,675],{"@type":290,"@id":628,"isPartOf":629,"author":630,"headline":127,"datePublished":612,"dateModified":614,"mainEntityOfPage":631,"wordCount":632,"publisher":633,"image":634,"thumbnailUrl":636,"articleSection":637,"inLanguage":366},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fserver-side-tagging-gtm\u002F#article",{"@id":603},{"name":4,"@id":357},{"@id":603},1419,{"@id":361},{"@id":635},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fserver-side-tagging-gtm\u002F#primaryimage","https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F11\u002Fmilad-fakurian-bMSA5-tLFao-unsplash.jpg?auto=compress%2Cformat&q=90",[136],{"@type":368,"@id":603,"url":603,"name":127,"isPartOf":639,"primaryImageOfPage":640,"image":641,"thumbnailUrl":636,"datePublished":612,"dateModified":614,"description":539,"breadcrumb":642,"inLanguage":375},{"@id":370},{"@id":635},{"@id":635},{"@id":643},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fserver-side-tagging-gtm\u002F#breadcrumb",{"@type":377,"inLanguage":366,"@id":635,"url":636,"contentUrl":636,"width":378,"height":645},1823,{"@type":381,"@id":643,"itemListElement":647},[648,649,650,652],{"@type":384,"position":385,"name":186,"item":386},{"@type":384,"position":388,"name":239,"item":389},{"@type":384,"position":391,"name":136,"item":651},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fcategory\u002Fmeasurement\u002F",{"@type":384,"position":394,"name":127},{"@type":396,"@id":370,"url":386,"name":318,"description":5,"publisher":654,"inLanguage":366,"potentialAction":655},{"@id":398},{"@type":400,"target":656,"query-input":404},{"@type":402,"urlTemplate":403},{"@type":406,"@id":398,"name":318,"url":386,"logo":658,"image":412,"sameAs":659,"@context":417,"email":418,"foundingDate":419,"legalName":420,"vatID":421,"description":422,"telephone":423,"address":660,"location":661,"areaServed":671},{"@type":377,"inLanguage":366,"@id":408,"url":409,"contentUrl":409,"width":410,"height":411,"caption":318},[414,415,416],{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},[662,665,668],{"@type":433,"name":434,"geo":663,"address":664,"photo":412,"telephone":440},{"@type":436,"latitude":437,"longitude":438},{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},{"@type":433,"name":442,"geo":666,"address":667,"photo":451,"telephone":452},{"@type":436,"latitude":444,"longitude":445},{"@type":425,"streetAddress":447,"addressLocality":448,"addressRegion":449,"postalCode":450,"addressCountry":430},{"@type":433,"name":454,"geo":669,"address":670},{"@type":436,"latitude":456,"longitude":457},{"@type":425,"streetAddress":459,"addressLocality":460,"addressRegion":461,"postalCode":462,"addressCountry":463},[672,673,674],{"@type":466,"name":467},{"@type":466,"name":469},{"@type":466,"name":463},{"@type":472,"@id":473,"name":4,"image":676,"description":6,"sameAs":677,"url":200},{"@type":377,"inLanguage":366,"@id":475,"url":8,"contentUrl":8,"caption":4},[196],{"self":679,"collection":684,"about":686,"author":688,"replies":690,"version-history":693,"predecessor-version":697,"wp:featuredmedia":701,"wp:attachment":704,"wp:term":707,"curies":714},[680],{"href":681,"targetHints":682},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F25590",{"allow":683},[483,484,485,486,487],[685],{"href":490},[687],{"href":493},[689],{"embeddable":108,"href":496},[691],{"embeddable":108,"href":692},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcomments?post=25590",[694],{"count":695,"href":696},18,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F25590\u002Frevisions",[698],{"id":699,"href":700},32725,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F25590\u002Frevisions\u002F32725",[702],{"embeddable":108,"href":703},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia\u002F25622",[705],{"href":706},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia?parent=25590",[708,710,712],{"taxonomy":516,"embeddable":108,"href":709},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcategories?post=25590",{"taxonomy":519,"embeddable":108,"href":711},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcollections?post=25590",{"taxonomy":180,"embeddable":108,"href":713},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcoauthors?post=25590",[715],{"name":525,"href":526,"templated":108},{"id":717,"date":718,"date_gmt":718,"guid":719,"modified":141,"modified_gmt":141,"slug":721,"status":19,"type":113,"link":140,"title":722,"content":723,"excerpt":725,"author":213,"featured_media":726,"comment_status":20,"ping_status":20,"sticky":211,"template":5,"format":215,"meta":727,"categories":729,"collections":731,"coauthors":732,"class_list":733,"acf":736,"head_links":737,"breadcrumbs":738,"url_path":140,"search_description":743,"featured_image_urls":744,"authors":776,"type_nicename":290,"reading_minutes":142,"category":788,"yoast_title":728,"yoast_meta":791,"yoast_json_ld":817,"_links":870},31394,"2024-12-19T17:20:33",{"rendered":720},"\u002F?p=31394","understanding-the-workings-of-bayesian-ab-testing",{"rendered":138},{"rendered":724,"protected":211},"\n\u003Cp>In digital marketing,\u003Ca rel=\"noreferrer noopener\" href=\"\u002Fdigital-experience\u002Fcro\u002Fab-testing\u002F\"> A\u002FB testing\u003C\u002Fa> is a technique for comparing two versions of a variable, such as a landing page layout or ad creative, to determine which version performs best with the target audience. The audience is randomly divided into two groups: one group is exposed to version A (the control), and the other to version B (the variant). Key performance indicators (KPI’s), such as conversions, are then tracked to evaluate each version’s performance.\u003C\u002Fp>\n\n\n\n\u003Cp>However, simply observing which version has a higher number of conversions doesn’t guarantee that the observed difference is due to the change itself. Random fluctuations or sampling variability can cause apparent differences that may not be meaningful. So, how can we be confident that any observed difference reflects a true effect rather than just chance? This is where statistics and \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fdigital-experience\u002Fcro\u002F\">data science\u003C\u002Fa> come into play to help quantify the uncertainty around our results and provide a rigorous basis for decision-making.\u003C\u002Fp>\n\n\n\n\u003Cdiv class=\"table-of-contents\">\u003Cul>\u003Cli class=\"heading-h2\">\u003Ca href=\"#the-bayesian-approach-to-a-b-testing\">The Bayesian approach to A\u002FB testing\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#frequentist-approach\">Frequentist approach\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#bayesian-approach\">Bayesian approach\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#bayesian-a-b-testing-implementation-in-cro\">Bayesian A\u002FB Testing Implementation in CRO\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#define-your-hypothesis\">Define your hypothesis\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#collect-and-interpret-your-data\">Collect and Interpret your data\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#choose-your-prior-conversion-rates\">Choose your prior conversion rates\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#calculate-the-posterior-conversion-rates\">Calculate the posterior conversion rates\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#calculate-the-probability-of-superiority\">Calculate the probability of superiority\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#calculate-the-conversion-rate-uplift\">Calculate the conversion rate uplift\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#interpret-your-results\">Interpret your results\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#take-action\">Take action\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Ful>\u003C\u002Fdiv>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"the-bayesian-approach-to-a-b-testing\">The Bayesian approach to A\u002FB testing\u003C\u002Fh2>\n\n\n\n\u003Cp>Statistics provides a framework for quantifying uncertainty, allowing us to make informed decisions even in the presence of incomplete information. This uncertainty is quantified through probability, giving us a way to understand the \u003Cstrong>likelihood\u003C\u002Fstrong> of different outcomes based on available data.\u003C\u002Fp>\n\n\n\n\u003Cp>In \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fbayesian-statistics-in-media-mix-modelling-a-real-world-analogy\u002F\">Bayesian statistics\u003C\u002Fa>, probability is viewed as a measure of certainty in a particular outcome and \u003Cstrong>prior\u003C\u002Fstrong> knowledge or beliefs are incorporated into the analysis. In Bayesian A\u002FB testing, the process begins with a prior belief (or distribution) about the effectiveness of each version (A and B), which might be informed by historical data, expert opinion, or previous experiments. As new data from the A\u002FB test is collected, the Bayesian approach updates this prior belief using Bayes’ theorem, resulting in a \u003Cstrong>posterior\u003C\u002Fstrong> probability distribution that reflects both the prior knowledge and the new evidence. This approach is particularly useful because it allows you to continuously refine your estimates as more data becomes available, leading to probability-based confidence in which version is likely to perform better.\u003C\u002Fp>\n\n\n\n\u003Cp>In A\u002FB testing, both Bayesian and Frequentist approaches serve distinct purposes:\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"frequentist-approach\">Frequentist approach\u003C\u002Fh3>\n\n\n\n\u003Cp>This approach is often valued for its objectivity and clear decision criteria, such as achieving \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fstatistical-significance-ppc\u002F\">statistical significance\u003C\u002Fa> at a specific \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fdigital-marketing-glossary\u002F#digital-marketing-terms-p-t\">p-value\u003C\u002Fa> threshold. However, it requires a fixed sample size and doesn’t directly tell us the probability that one version is better than the other. It only indicates whether the observed difference is likely due to chance. \u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"bayesian-approach\">Bayesian approach\u003C\u002Fh3>\n\n\n\n\u003Cp>This approach, by contrast, provides a direct probability that one version is better than the other, allowing for a more flexible interpretation of the results and enabling dynamic decision-making as data accumulates. This flexibility is particularly advantageous in digital marketing, where strategies often need to be adjusted in real-time based on rapidly changing user behaviours and market conditions.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"bayesian-a-b-testing-implementation-in-cro\">Bayesian A\u002FB Testing Implementation in CRO\u003C\u002Fh2>\n\n\n\n\u003Cp>The remainder of this blog will walk you through how to implement your own Bayesian A\u002FB test. While there are many \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fthe-best-a-b-testing-tools\u002F\">A\u002FB testing tools\u003C\u002Fa> available that can quickly provide probabilities, uplift estimates, and other key metrics, understanding the internal workings behind these tools is crucial. Knowing how concepts like posterior probabilities, credible intervals, and Bayesian inference work enables you to interpret results more effectively and make smarter decisions. It also helps you recognize the limitations of automated tools, ensuring that your strategies are based on sound analysis rather than blind trust in numbers.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"define-your-hypothesis\">Define your hypothesis\u003C\u002Fh3>\n\n\n\n\u003Cp>The first step in any A\u002FB test is to define your \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fwhere-to-start-with-a-testing-hypothesis\u002F\">hypothesis\u003C\u002Fa>, this is an experiment after all.\u003C\u002Fp>\n\n\n\n\u003Cp>In the context of \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fdigital-experience\u002Fcro\u002F\">conversion rate optimisation\u003C\u002Fa> (CRO), the \u003Cstrong>null hypothesis (H\u003C\u002Fstrong>\u003Cstrong>\u003Csub>0\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong>)\u003C\u002Fstrong> and the \u003Cstrong>alternative hypothesis (H₁)\u003C\u002Fstrong> are defined as follows:\u003C\u002Fp>\n\n\n\n\u003Cul class=\"wp-block-list\">\n\u003Cli>\u003Cstrong>Null Hypothesis (H₀):\u003C\u002Fstrong> There is no difference in the performance between version A (the control) and version B (the variant). Any observed difference in conversion rates is purely due to random chance. Mathematically, this is expressed as:\u003C\u002Fli>\n\u003C\u002Ful>\n\n\n\n\u003Cp class=\"has-text-align-center\">\u003Cstrong>H\u003C\u002Fstrong>\u003Cstrong>\u003Csub>0\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong>:\u003C\u002Fstrong> \u003Csub> \u003C\u002Fsub>CR\u003Csub>A\u003C\u002Fsub> = CR\u003Csub>B\u003C\u002Fsub>\u003C\u002Fp>\n\n\n\n\u003Cul class=\"wp-block-list\">\n\u003Cli>\u003Cstrong>Alternative Hypothesis (H₁):\u003C\u002Fstrong> There is a statistically significant difference in performance between version A and version B. This means that any observed difference is likely due to the actual effect of the changes made in version B, rather than random chance. This can be one-sided (suggesting that one version performs better than the other) or two-sided (suggesting that the two versions perform differently, without specifying which is better). A two-sided alternative hypothesis can be mathematically expressed as:\u003C\u002Fli>\n\u003C\u002Ful>\n\n\n\n\u003Cp class=\"has-text-align-center\">\u003Cstrong>H\u003Csub>1\u003C\u002Fsub>:\u003C\u002Fstrong> CR\u003Csub>A\u003C\u002Fsub> ≠ CR\u003Csub>B\u003C\u002Fsub>\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"collect-and-interpret-your-data\">Collect and Interpret your data\u003C\u002Fh3>\n\n\n\n\u003Cp>Randomly assign users from the target audience to one of groups A or B and collect the relevant conversion data. Your dataset should look something like the following:\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-table\">\u003Ctable>\u003Ctbody>\u003Ctr>\u003Ctd>\u003C\u002Ftd>\u003Ctd>\u003Cstrong>Sample Size\u003C\u002Fstrong>\u003C\u002Ftd>\u003Ctd>\u003Cstrong>Number of Conversions\u003C\u002Fstrong>\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>\u003Cstrong>Group A\u003C\u002Fstrong>\u003C\u002Ftd>\u003Ctd>2100\u003C\u002Ftd>\u003Ctd>58\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>\u003Cstrong>Group B\u003C\u002Fstrong>\u003C\u002Ftd>\u003Ctd>2155\u003C\u002Ftd>\u003Ctd>79\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Ffigure>\n\n\n\n\u003Cp>To interpret our data, we need to think about the underlying likelihood of our data.\u003C\u002Fp>\n\n\n\n\u003Cp>We know that a random user from group A will convert with an unknown conversion rate CR\u003Csub>A\u003C\u002Fsub>. As this outcome is binary (convert = 1 or not convert  = 0), the event that a random user (\u003Cstrong>X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>1A\u003C\u002Fsub>\u003C\u002Fstrong>) in group A will convert, given the conversion rate CR\u003Csub>A\u003C\u002Fsub>, can be described using a \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fmathworld.wolfram.com\u002FBernoulliDistribution.html\">Bernoulli distribution\u003C\u002Fa>:\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">\u003Cstrong>X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>1A\u003C\u002Fsub>\u003C\u002Fstrong> \u003Cstrong>| \u003C\u002Fstrong>CR\u003Csub>A\u003C\u002Fsub> ~ \u003Cstrong>Bernoulli\u003C\u002Fstrong>(CR\u003Csub>A\u003C\u002Fsub>)\u003C\u002Fp>\n\n\n\n\u003Cp>Given we have N\u003Csub>A\u003C\u002Fsub> users in group A, the \u003Cstrong>likelihood\u003C\u002Fstrong> of our data for group A is given by the sum of N\u003Csub>A\u003C\u002Fsub> Bernoulli trials.\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">(\u003Cstrong>X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>1A\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> \u003C\u002Fstrong>+\u003Cstrong> X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>2A\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> \u003C\u002Fstrong>+ … +\u003Cstrong> X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>NA\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> | \u003C\u002Fstrong>CR\u003Csub>A\u003C\u002Fsub>)\u003C\u002Fp>\n\n\n\n\u003Cp>Which follows the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fmathworld.wolfram.com\u002FBinomialDistribution.html\">Binomial distribution\u003C\u002Fa>:\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">(\u003Cstrong>X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>1A\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> \u003C\u002Fstrong>+\u003Cstrong> X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>2A\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> \u003C\u002Fstrong>+ … +\u003Cstrong> X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>NA\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> | \u003C\u002Fstrong>CR\u003Csub>A\u003C\u002Fsub>) ~ \u003Cstrong>Binomial\u003C\u002Fstrong>(N\u003Csub>A\u003C\u002Fsub>,CR\u003Csub>A\u003C\u002Fsub>)\u003C\u002Fp>\n\n\n\n\u003Cp>Through similar analysis, the \u003Cstrong>likelihood\u003C\u002Fstrong> of our data for group B is given by:\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">(\u003Cstrong>X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>1B\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> \u003C\u002Fstrong>+\u003Cstrong> X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>2B\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> \u003C\u002Fstrong>+ … +\u003Cstrong> X\u003C\u002Fstrong>\u003Cstrong>\u003Csub>NB\u003C\u002Fsub>\u003C\u002Fstrong>\u003Cstrong> | \u003C\u002Fstrong>CR\u003Csub>B\u003C\u002Fsub>) ~ \u003Cstrong>Binomial\u003C\u002Fstrong>(N\u003Csub>B\u003C\u002Fsub>,CR\u003Csub>B\u003C\u002Fsub>)\u003C\u002Fp>\n\n\n\n\u003Cp>Now we’ve collected our data and determined the likelihood distributions of both groups follow a binomial distribution, the next step is to choose your prior.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1211\" height=\"611\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fbinomial-likelihood-distribution.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-31404\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"choose-your-prior-conversion-rates\">Choose your prior conversion rates\u003C\u002Fh3>\n\n\n\n\u003Cp>A \u003Cstrong>non-informative prior\u003C\u002Fstrong> is used when we have little to no prior knowledge about the likely outcomes. It ensures the data will primarily drive the results rather than any preconceived assumptions. An \u003Cstrong>informative prior\u003C\u002Fstrong>, on the other hand, incorporates existing knowledge or beliefs about the expected outcome based on past experiments, domain expertise, or historical data. Informative priors can risk introducing unwanted bias if the prior is inaccurate. \u003C\u002Fp>\n\n\n\n\u003Cp>In the case where we have very little (if any) knowledge of expected outcomes, a commonly used non-informative prior is a \u003Cstrong>Beta\u003C\u002Fstrong>(1,1) distribution, a special case of the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fmathworld.wolfram.com\u002FBetaDistribution.html\">Beta distribution\u003C\u002Fa>.\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">CR\u003Csub>A\u003C\u002Fsub> ~ \u003Cstrong>Beta\u003C\u002Fstrong>(1,1)\u003C\u002Fp>\n\n\n\n\u003Cp>This prior assumes the true conversion rate exists anywhere between 0 and 1, with equal probability.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1211\" height=\"611\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fbeta-distribution.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-31407\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"calculate-the-posterior-conversion-rates\">Calculate the posterior conversion rates\u003C\u002Fh3>\n\n\n\n\u003Cp>In Bayesian statistics, calculating the posterior distribution is essential to updating beliefs about a hypothesis given observed data, and this is achieved using Bayes’ theorem. Here, Bayes’ theorem combines prior knowledge (the prior distribution) with new data (through the likelihood) to compute the posterior distribution, which represents an updated probability of the hypothesis.\u003C\u002Fp>\n\n\n\n\u003Cp>For the full computation of the posterior, we need the prior, the likelihood, and the marginal probability of the observed data. The marginal probability (also called the evidence) is typically challenging to calculate directly since it involves integrating all possible values of the parameter, which can be computationally expensive or infeasible in complex models such as \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fmedia-mix-modelling\u002F\">Marketing Mix Models (MMM)\u003C\u002Fa>. Therefore, in many cases, we rely on advanced sampling techniques, such as Markov Chain Monte Carlo (MCMC), to approximate the posterior.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"834\" height=\"524\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fmarkov-chain-monte-carlo.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-31406\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Cp>However, in Bayesian A\u002FB testing, if we use a non-informative prior, such as a \u003Cstrong>Beta\u003C\u002Fstrong>(1,1) distribution, our computation simplifies significantly. This prior (and any Beta prior) is \u003Cstrong>conjugate\u003C\u002Fstrong> to the binomial likelihood, meaning that when combined with binomial data, it produces a posterior that is also in the \u003Cstrong>Beta distribution family\u003C\u002Fstrong>. Conjugate priors are particularly useful in Bayesian analysis because they allow for an exact analytical solution of the posterior without the need for sampling methods. \u003C\u002Fp>\n\n\n\n\u003Cp>By choosing a Beta prior with parameters that are either non-informative or reflective of prior knowledge, we can obtain an updated Beta posterior that directly incorporates our A\u002FB testing results, making the Bayesian analysis more straightforward and computationally efficient.\u003C\u002Fp>\n\n\n\n\u003Cp>After working through the Mathematics, we obtain the following posterior distribution for CR\u003Csub>A\u003C\u002Fsub>:\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">(CR\u003Csub>A\u003C\u002Fsub> | \u003Cstrong>X\u003Csub>1A\u003C\u002Fsub> \u003C\u002Fstrong>+\u003Cstrong> X\u003Csub>2A\u003C\u002Fsub> \u003C\u002Fstrong>+ … +\u003Cstrong> X\u003Csub>NA\u003C\u002Fsub>\u003C\u002Fstrong>) ~\u003Cstrong> Beta\u003C\u002Fstrong>(\u003Cstrong>Σx\u003Csub>A\u003C\u002Fsub>\u003C\u002Fstrong>+1, N\u003Csub>A\u003C\u002Fsub>–\u003Cstrong>Σ\u003C\u002Fstrong>x\u003Cstrong>\u003Csub>A\u003C\u002Fsub>\u003C\u002Fstrong>+1)\u003C\u002Fp>\n\n\n\n\u003Cp>Where\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">\u003Cstrong>Σx\u003Csub>A\u003C\u002Fsub> \u003C\u002Fstrong>is the total number of observed conversions in group A\u003C\u002Fp>\n\n\n\n\u003Cp>Through similar analysis, the \u003Cstrong>posterior\u003C\u002Fstrong> distribution for CR\u003Csub>B\u003C\u002Fsub> is given by:\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">(CR\u003Csub>B\u003C\u002Fsub> | \u003Cstrong>X\u003Csub>1B\u003C\u002Fsub> \u003C\u002Fstrong>+\u003Cstrong> X\u003Csub>2B\u003C\u002Fsub> \u003C\u002Fstrong>+ … +\u003Cstrong> X\u003Csub>NB\u003C\u002Fsub>\u003C\u002Fstrong>) ~\u003Cstrong> Beta\u003C\u002Fstrong>(\u003Cstrong>Σx\u003Csub>B\u003C\u002Fsub>\u003C\u002Fstrong>+1, N\u003Csub>B\u003C\u002Fsub>–\u003Cstrong>Σx\u003Csub>B\u003C\u002Fsub>\u003C\u002Fstrong>+1)\u003C\u002Fp>\n\n\n\n\u003Cp>Where\u003C\u002Fp>\n\n\n\n\u003Cp class=\"has-text-align-center\">\u003Cstrong>Σx\u003Csub>B\u003C\u002Fsub> \u003C\u002Fstrong>is the total number of observed conversions in group B\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1010\" height=\"554\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fposterior-conversion-rate-for-versions-a-b.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-31409\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"calculate-the-probability-of-superiority\">Calculate the probability of superiority\u003C\u002Fh3>\n\n\n\n\u003Cp>The \u003Cstrong>probability of superiority\u003C\u002Fstrong> quantifies the likelihood that one group (B) performs better than another (A). It is calculated by taking the difference between the posterior distributions of both versions and then determining the proportion of the resulting distribution that is greater than zero. This area under the density curve where the difference is positive represents the confidence that variant B is superior to control A.\u003C\u002Fp>\n\n\n\n\u003Cp>For the visual below, the probability superiority is 95.2%.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1001\" height=\"553\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fprobability-of-superiority.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-31408\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"calculate-the-conversion-rate-uplift\">Calculate the conversion rate uplift\u003C\u002Fh3>\n\n\n\n\u003Cp>The \u003Cstrong>relative conversion rate uplift\u003C\u002Fstrong> is given by the following:\u003C\u002Fp>\n\n\n\n\u003Cp>CR\u003Csub>B\u003C\u002Fsub> – CR\u003Csub>A\u003C\u002Fsub> \u002F CR\u003Csub>A\u003C\u002Fsub>\u003C\u002Fp>\n\n\n\n\u003Cp>As we have posterior distributions for both CR\u003Csub>A\u003C\u002Fsub> and CR\u003Csub>B\u003C\u002Fsub>, we can now estimate the posterior  relative conversion rate uplift. This is highly beneficial because the posterior provides \u003Cstrong>Highest Density Intervals (HDIs)\u003C\u002Fstrong> for the relative conversion rate uplift, offering a clear range within which the true uplift is most likely to fall. Unlike traditional methods that give a binary “significant or not” result, HDIs enable you to quantify uncertainty and understand the potential variability in uplift.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1015\" height=\"553\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fposterior-relative-conversion-rate-uplift.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-31405\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"interpret-your-results\">Interpret your results\u003C\u002Fh3>\n\n\n\n\u003Cp>The \u003Cstrong>probability of superiority\u003C\u002Fstrong> gives you the probability that one version is better than the other. In this case, variant B has a 95.2% probability of being superior to control A, meaning we have evidence to reject the null hypothesis and can be confident that version B is likely the better choice.\u003C\u002Fp>\n\n\n\n\u003Cp>The \u003Cstrong>posterior conversion rates\u003C\u002Fstrong> can provide valuable insights too. You can use them to see the range of possible true conversion rates for each version. This range reflects uncertainty, so a wider interval means your data might not provide a clear answer, while a narrow interval boosts confidence. In this case, both distributions appear to have similar dispersion, but variant B is slightly wider. Although both are very similar in terms of uncertainty, we’re slightly less confident in the true conversion rate of variant B than control A.\u003C\u002Fp>\n\n\n\n\u003Cp>Finally, the \u003Cstrong>posterior conversion rate uplift\u003C\u002Fstrong> shows how much better one version might perform, along with credible intervals that capture its uncertainty. In this case, the mean conversion rate uplift is 34.3%, meaning we estimate variant B will generate 34.3% more conversions on average than variant A. By analysing the credible interval, we can infer that variant B will generate between -4.56% and 83.68% more conversions with variant A with 95% probability. As this interval is very wide, we must take extreme caution when implementing variant B as standard.\u003C\u002Fp>\n\n\n\n\u003Cp>When interpreting the results of your Bayesian A\u002FB test, focus on the story your data is telling. Instead of just focusing on the average values, pay attention to credible intervals – they help you weigh the potential benefits of a change against the risks of uncertainty. By understanding both the insights and limitations of your results, you’ll make smarter, more confident decisions for your business.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"take-action\">Take action\u003C\u002Fh3>\n\n\n\n\u003Cp>With the results of your Bayesian A\u002FB test in hand, it’s time to take action. The analysis shows that variant B has a 95.2% probability of being superior to control A, with an estimated average uplift of 34.3% in conversion rate. This strong evidence suggests implementing variant B as the new standard is a smart choice. Use the insights from the credible intervals (-4.56% to 83.68%) to set realistic expectations for performance improvements, account for the potential risks and plan your strategy accordingly. \u003C\u002Fp>\n\n\n\n\u003Cp>Beyond simply adopting variant B, leverage what you’ve learned to iterate further by identifying the elements of variant B that drove the uplift and test additional refinements. Bayesian A\u002FB testing isn’t just about finding “winners”, it’s about continuously learning and improving to drive sustainable growth for your business.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002F\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"3840\" height=\"540\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F04\u002Ff0sKYLsk-Measurement_Banner_01.jpg?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-27925\"\u002F>\u003C\u002Fa>\u003C\u002Ffigure>\n",{"rendered":139,"protected":211},31413,{"_yoast_wpseo_focuskw":5,"_yoast_wpseo_title":728,"_yoast_wpseo_metadesc":5,"inline_featured_image":211,"footnotes":5},"Understanding the workings of Bayesian A\u002FB testing | Impression",[730,219],280,[],[223],[734,113,226,227,228,229,230,735,231],"post-31394","category-data-engineering",[],[],[739,740,741,742],{"path":185,"name":186},{"path":238,"name":239},{"path":147,"name":148},{"path":194,"name":138},"In digital marketing, A\u002FB testing is a technique for comparing two versions of a variable, such as a landing page layout or ad creative, to determine which version performs best with the target audience. The audience is randomly divided into two groups: one group is exposed to version A (the control), and the other to […]",{"thumbnail":745,"medium":746,"medium_large":748,"large":750,"1536x1536":752,"2048x2048":754,"post-thumbnail":756,"12_col":758,"default_landscape":760,"default_portrait":762,"6_col_and_margin_short":764,"11_col_and_margin":766,"9_col_rectangle":768,"11_col":770,"staff_image_696_600":772,"rectangular_thumbnail":774},{"url":143,"width":244,"height":244},{"url":747,"width":70,"height":70},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=200&q=90&w=133",{"url":749,"width":73,"height":23},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1152&q=90&w=768",{"url":751,"width":251,"height":252},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1000&q=90&w=1892",{"url":753,"width":79,"height":79},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1536&q=90&w=1024",{"url":755,"width":82,"height":82},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=2048&q=90&w=1365",{"url":757,"width":85,"height":86},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1180&q=90&w=1860",{"url":759,"width":88,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2880",{"url":761,"width":91,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=800",{"url":763,"width":65,"height":91},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=800&q=90&w=600",{"url":765,"width":85,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=1860",{"url":767,"width":95,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=3080",{"url":769,"width":97,"height":98},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1400&q=90&w=2150",{"url":771,"width":100,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2640",{"url":773,"width":64,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=696",{"url":775,"width":103,"height":104},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=300&q=90&w=750",[777],{"id":213,"path":279,"display_name":4,"job_title":11,"description":6,"avatar_url":122,"acf":778},{"full_bio":10,"job_title":11,"specialist_service_links":779,"photo":786},[780,782,784],{"page":781,"link_text":29},{"ID":15,"post_author":16,"post_date":17,"post_date_gmt":17,"post_content":5,"post_title":18,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":21,"to_ping":5,"pinged":5,"post_modified":22,"post_modified_gmt":22,"post_content_filtered":5,"post_parent":23,"guid":24,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":28},{"page":783,"link_text":34},{"ID":32,"post_author":16,"post_date":33,"post_date_gmt":33,"post_content":5,"post_title":34,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":35,"to_ping":5,"pinged":5,"post_modified":36,"post_modified_gmt":36,"post_content_filtered":5,"post_parent":37,"guid":38,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":39},{"page":785,"link_text":44},{"ID":42,"post_author":16,"post_date":43,"post_date_gmt":43,"post_content":5,"post_title":44,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":45,"to_ping":5,"pinged":5,"post_modified":46,"post_modified_gmt":46,"post_content_filtered":5,"post_parent":37,"guid":47,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":48},{"ID":50,"id":50,"title":51,"filename":52,"filesize":53,"url":54,"link":55,"alt":5,"author":56,"description":5,"caption":5,"name":57,"status":58,"uploaded_to":23,"date":59,"modified":59,"menu_order":23,"mime_type":60,"type":61,"subtype":62,"icon":63,"width":64,"height":65,"sizes":787},{"thumbnail":67,"thumbnail-width":68,"thumbnail-height":68,"medium":69,"medium-width":70,"medium-height":71,"medium_large":72,"medium_large-width":73,"medium_large-height":74,"large":75,"large-width":76,"large-height":77,"1536x1536":78,"1536x1536-width":79,"1536x1536-height":80,"2048x2048":81,"2048x2048-width":82,"2048x2048-height":83,"post-thumbnail":84,"post-thumbnail-width":85,"post-thumbnail-height":86,"12_col":87,"12_col-width":88,"12_col-height":89,"default_landscape":90,"default_landscape-width":91,"default_landscape-height":65,"default_portrait":92,"default_portrait-width":65,"default_portrait-height":91,"6_col_and_margin_short":93,"6_col_and_margin_short-width":85,"6_col_and_margin_short-height":89,"11_col_and_margin":94,"11_col_and_margin-width":95,"11_col_and_margin-height":89,"9_col_rectangle":96,"9_col_rectangle-width":97,"9_col_rectangle-height":98,"11_col":99,"11_col-width":100,"11_col-height":89,"staff_image_696_600":101,"staff_image_696_600-width":64,"staff_image_696_600-height":65,"rectangular_thumbnail":102,"rectangular_thumbnail-width":103,"rectangular_thumbnail-height":104},[789,790],{"id":730,"name":148,"path":147},{"id":219,"name":125,"path":124},[792,793,794,796,797,798,799,800,801,802,803,805,807,809,810,811,812,813,814,815,816],{"property":296,"content":728},{"name":300,"content":301},{"property":199,"content":795},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Funderstanding-the-workings-of-bayesian-ab-testing\u002F",{"name":305,"content":306},{"property":308,"content":309},{"property":311,"content":728},{"property":313,"content":743},{"property":315,"content":795},{"property":317,"content":318},{"property":320,"content":321},{"property":323,"content":804},"2024-12-19T17:20:33+00:00",{"property":326,"content":806},"2024-12-19T17:20:34+00:00",{"property":329,"content":808},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fopen-graph-inner-workings-ab-testing.png?auto=compress%2Cformat&q=90",{"property":332,"content":333},{"property":335,"content":336},{"property":338,"content":60},{"property":180,"content":4},{"property":341,"content":342},{"property":344,"content":808},{"property":346,"content":347},{"property":349,"content":347},{"@context":351,"@graph":818},[819,830,836,838,845,849,867],{"@type":290,"@id":820,"isPartOf":821,"author":822,"headline":138,"datePublished":804,"dateModified":806,"mainEntityOfPage":823,"wordCount":824,"publisher":825,"image":826,"thumbnailUrl":828,"articleSection":829,"inLanguage":366},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Funderstanding-the-workings-of-bayesian-ab-testing\u002F#article",{"@id":795},{"name":4,"@id":357},{"@id":795},2003,{"@id":361},{"@id":827},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Funderstanding-the-workings-of-bayesian-ab-testing\u002F#primaryimage","https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F12\u002Fhans-isaacson-O6uM94iMyVk-unsplash.jpg?auto=compress%2Cformat&q=90",[148,125],{"@type":368,"@id":795,"url":795,"name":728,"isPartOf":831,"primaryImageOfPage":832,"image":833,"thumbnailUrl":828,"datePublished":804,"dateModified":806,"breadcrumb":834,"inLanguage":375},{"@id":370},{"@id":827},{"@id":827},{"@id":835},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Funderstanding-the-workings-of-bayesian-ab-testing\u002F#breadcrumb",{"@type":377,"inLanguage":366,"@id":827,"url":828,"contentUrl":828,"width":837,"height":88},1920,{"@type":381,"@id":835,"itemListElement":839},[840,841,842,844],{"@type":384,"position":385,"name":186,"item":386},{"@type":384,"position":388,"name":239,"item":389},{"@type":384,"position":391,"name":148,"item":843},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fcategory\u002Fdata-engineering\u002F",{"@type":384,"position":394,"name":138},{"@type":396,"@id":370,"url":386,"name":318,"description":5,"publisher":846,"inLanguage":366,"potentialAction":847},{"@id":398},{"@type":400,"target":848,"query-input":404},{"@type":402,"urlTemplate":403},{"@type":406,"@id":398,"name":318,"url":386,"logo":850,"image":412,"sameAs":851,"@context":417,"email":418,"foundingDate":419,"legalName":420,"vatID":421,"description":422,"telephone":423,"address":852,"location":853,"areaServed":863},{"@type":377,"inLanguage":366,"@id":408,"url":409,"contentUrl":409,"width":410,"height":411,"caption":318},[414,415,416],{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},[854,857,860],{"@type":433,"name":434,"geo":855,"address":856,"photo":412,"telephone":440},{"@type":436,"latitude":437,"longitude":438},{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},{"@type":433,"name":442,"geo":858,"address":859,"photo":451,"telephone":452},{"@type":436,"latitude":444,"longitude":445},{"@type":425,"streetAddress":447,"addressLocality":448,"addressRegion":449,"postalCode":450,"addressCountry":430},{"@type":433,"name":454,"geo":861,"address":862},{"@type":436,"latitude":456,"longitude":457},{"@type":425,"streetAddress":459,"addressLocality":460,"addressRegion":461,"postalCode":462,"addressCountry":463},[864,865,866],{"@type":466,"name":467},{"@type":466,"name":469},{"@type":466,"name":463},{"@type":472,"@id":473,"name":4,"image":868,"description":6,"sameAs":869,"url":200},{"@type":377,"inLanguage":366,"@id":475,"url":8,"contentUrl":8,"caption":4},[196],{"self":871,"collection":876,"about":878,"author":880,"replies":882,"version-history":885,"predecessor-version":889,"wp:featuredmedia":893,"wp:attachment":896,"wp:term":899,"curies":906},[872],{"href":873,"targetHints":874},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F31394",{"allow":875},[483,484,485,486,487],[877],{"href":490},[879],{"href":493},[881],{"embeddable":108,"href":496},[883],{"embeddable":108,"href":884},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcomments?post=31394",[886],{"count":887,"href":888},9,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F31394\u002Frevisions",[890],{"id":891,"href":892},31412,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F31394\u002Frevisions\u002F31412",[894],{"embeddable":108,"href":895},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia\u002F31413",[897],{"href":898},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia?parent=31394",[900,902,904],{"taxonomy":516,"embeddable":108,"href":901},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcategories?post=31394",{"taxonomy":519,"embeddable":108,"href":903},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcollections?post=31394",{"taxonomy":180,"embeddable":108,"href":905},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcoauthors?post=31394",[907],{"name":525,"href":526,"templated":108},{"id":909,"date":910,"date_gmt":910,"guid":911,"modified":153,"modified_gmt":153,"slug":913,"status":19,"type":113,"link":152,"title":914,"content":915,"excerpt":917,"author":213,"featured_media":918,"comment_status":20,"ping_status":20,"sticky":211,"template":5,"format":215,"meta":919,"categories":922,"collections":923,"coauthors":924,"class_list":925,"acf":927,"head_links":928,"breadcrumbs":929,"url_path":152,"search_description":921,"featured_image_urls":934,"authors":966,"type_nicename":290,"reading_minutes":154,"category":978,"yoast_title":920,"yoast_meta":981,"yoast_json_ld":1008,"_links":1060},30611,"2024-10-03T16:37:10",{"rendered":912},"\u002F?p=30611","granularity-marketing-mix-modelling",{"rendered":150},{"rendered":916,"protected":211},"\n\u003Cp>\u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fmedia-mix-modelling\u002F\">Marketing Mix Modeling\u003C\u002Fa> (MMM) is a statistical method used to measure the effectiveness of marketing activities and other factors on sales, helping you understand how your marketing efforts contribute to results – with a holistic methodology.\u003C\u002Fp>\n\n\n\n\u003Cdiv class=\"table-of-contents\">\u003Cul>\u003Cli class=\"heading-h2\">\u003Ca href=\"#identifying-when-your-data-is-too-granular\">Identifying when your data is too granular\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#increasing-granularity-without-added-complexity\">Increasing granularity without added complexity\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#increasing-granularity-with-incrementality-testing\">Increasing granularity with incrementality testing\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#increasing-granularity-with-sub-modelling\">Increasing granularity with sub-modelling\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#which-approach-should-you-take\">Which approach should you take?\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Ful>\u003C\u002Fdiv>\n\n\n\n\u003Cp>MMM enables you to optimise budget allocation by analysing the effectiveness of each marketing channel and identifying the point of diminishing returns through response curves. Response curves are used to visualise the relationship between marketing inputs and sales. For example, an MMM output might show that TV ads contribute 20% to sales but exceed saturation, while paid search drives 35% with room for growth. In this case, reducing TV ad spend and reallocating it to paid search could lead to a significant increase in overall sales. This data-driven adjustment improves your marketing mix, ensuring that your marketing spend delivers maximum impact.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1009\" height=\"552\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fpaid-social-response-curve.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-30618\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Cp>Traditionally, when conducting MMM, it was best practice to split your \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fdata-requirements-for-marketing-mix-modelling\u002F\">input data\u003C\u002Fa> into broader channels (like TV, Google ads, print or OOH) rather than dividing each channel into specific campaign types or funnel stages. This is because breaking it down too much can lead to an overly complex model which may overfit the data and become too tailored to past trends. When overfitting occurs, your MMM tends to lose its ability to generalise for future predictions, making it harder to interpret and less reliable.\u003C\u002Fp>\n\n\n\n\u003Cp>Overfitting results in poor forecasting ability, making the model unreliable for future media planning. Relying on such a model increases your risk of making bad decisions, such as over-investing in ineffective channels or underfunding high-performing ones, ultimately leading to inefficient budget allocation, wasted resources, and reduced ROI on marketing campaigns.  \u003C\u002Fp>\n\n\n\n\u003Cp>However, as marketers, we often wish to understand the impact of specific campaigns, targeting strategies, or funnel stages within each channel. This blog will show you how to identify when your MMM may be too complex and guide you through two ways of achieving more granular insights without jeopardising model complexity. By using advanced techniques and careful aggregation, it’s possible to maintain a balance between detailed insights and a manageable, reliable model that remains easy to interpret and useful for future decision-making.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"identifying-when-your-data-is-too-granular\">Identifying when your data is too granular\u003C\u002Fh2>\n\n\n\n\u003Cp>When assessing whether your MMM input data is too granular, you should consider if the level of detail being used is fit for purpose. While it’s tempting to break down data into specific campaigns, audience segments, or even individual ads, this can introduce an overwhelming number of variables into the model. The more granular the data, the more parameters the model has to estimate, increasing the likelihood that it will become overly complex and less capable of generalising beyond the dataset used for training.\u003C\u002Fp>\n\n\n\n\u003Cp>As the number of parameters increases, the risk of overfitting and instability grows. You should strive to strike a balance between capturing enough detail to \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fguide-to-media-effectiveness-measurement\u002F#media-mix-modelling\">generate actionable insights\u003C\u002Fa> and maintaining a model that’s simple enough to generalise well. Simplifying your input data without losing key insights can help keep the model efficient and reliable.\u003C\u002Fp>\n\n\n\n\u003Cp>Finding the balance between granularity and model simplicity involves using cross-validation techniques that go beyond simple in-sample performance metrics like \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fdigital-marketing-glossary\u002F#digital-marketing-terms-p-t\">R²\u003C\u002Fa>. While a high in-sample R² might seem like a good sign, it often indicates that the model is overfitting, capturing noise rather than true, predictive relationships. In-sample R² only reflects how well the model explains the historical data, but it doesn’t provide insight into how well the model will generalise to unseen data.\u003C\u002Fp>\n\n\n\n\u003Cp>A more robust approach is to use Leave-One-Out Cross-Validation (LOO-CV) or other \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fbuilding-trustworthy-media-insights-through-model-validation\u002F\">cross-validation methods\u003C\u002Fa>, which split the data into training and validation sets repeatedly to test the model’s ability to generalise. LOO-CV, in particular, helps assess how well the model performs on unseen data by leaving out one observation at a time, training the model on the rest, and then predicting the omitted data point. This technique provides a more reliable check against overfitting, ensuring that the model is capturing real patterns rather than noise.\u003C\u002Fp>\n\n\n\n\u003Cp>Another valuable method is \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing\u002F#hold-out-testing-for-the-synthetic-control\">hold-out testing\u003C\u002Fa>, where a portion of the data is set aside before training the model. After the model is built, its performance is evaluated on this hold-out set to simulate how well it would work in a real-world, future scenario. If the model performs well on the hold-out data, it’s a sign that the level of granularity and complexity is appropriate. If not, it may suggest that the model is too complex and overfitting to the training data, requiring simplification of the input data or a reduction in parameters.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"increasing-granularity-without-added-complexity\">Increasing granularity without added complexity\u003C\u002Fh2>\n\n\n\n\u003Cp>We can increase granularity without adding complexity by using techniques like \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fleveraging-incrementality-testing-to-optimise-media-mix-modelling\u002F#what-is-incrementality-testing\">incrementality testing\u003C\u002Fa> or sub-modelling, which allow for more detailed insights without overwhelming the model.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"increasing-granularity-with-incrementality-testing\">Increasing granularity with incrementality testing\u003C\u002Fh3>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"948\" height=\"280\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fcausal-impact-granularity.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-30616\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Cp>After running an initial channel-level MMM, you might find that a channel like Meta ads has a high return on ad spend (ROAS), but this insight doesn’t give you specific details on which campaigns or objectives are driving that impact. To expand these results, you can use incrementality testing to break down the contribution of specific campaigns or objectives within the Meta ads channel. \u003C\u002Fp>\n\n\n\n\u003Cp>For example, you could run a \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing\u002F#incrementality-test-designs\">controlled experiment\u003C\u002Fa> by temporarily pausing or adjusting spend on certain campaigns (or on top vs bottom-funnel efforts) to measure the change in overall performance. The difference between the baseline performance (without the specific campaign or funnel component) and the performance during the test period gives you an \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing\u002F#how-to-interpret-the-results-of-your-incrementality-test\">estimate\u003C\u002Fa> of that campaign’s or funnel stage’s incremental impact.\u003C\u002Fp>\n\n\n\n\u003Cp>Once the incrementality testing is complete, these results can be \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing\u002F#how-does-incrementality-testing-complement-marketing-mix-modelling\">integrated into your MMM\u003C\u002Fa> by attributing a portion of the channel’s overall contribution to the specific campaigns or funnel stages tested. For example, if the test shows that a particular retargeting campaign at the bottom of the funnel is responsible for 30% of the Meta ads sales, you can adjust the channel-level model to account for this breakdown. By repeating this process for other campaigns or funnel stages, you can expand the initial channel-level insights into more granular insights, helping to identify which parts of the funnel or campaigns are most effective.\u003C\u002Fp>\n\n\n\n\u003Cp>The key benefit of using incrementality testing to obtain more granular insights is that it provides a more accurate understanding of which campaigns or funnel stages are truly driving results, allowing for better optimization of marketing spend. It also avoids overcomplicating the initial MMM by keeping the channel-level model intact and layering more granular insights on top. However, there are limitations, such as the need for controlled environments and enough data to perform reliable tests. Additionally, incrementality testing can be time-consuming, and the results may not always be generalizable across all campaigns or stages of the marketing funnel.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" id=\"increasing-granularity-with-sub-modelling\">Increasing granularity with sub-modelling\u003C\u002Fh3>\n\n\n\u003Cdiv class=\"wp-block-image\">\n\u003Cfigure class=\"aligncenter size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"512\" height=\"388\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fgoogle-ad-contributions-sub-modelling.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-30617\"\u002F>\u003C\u002Ffigure>\u003C\u002Fdiv>\n\n\n\u003Cp>After running an initial channel-level MMM, you might want to expand the insights to understand how different campaigns or funnel stages within a channel contribute to overall performance. Another way to do this is by creating a sub-model that focuses specifically on campaign types or funnel levels, using data such as spend or impressions as input. For instance, if your MMM shows that Google ads drive significant sales, you could create a sub-model to determine how different campaign types within the Google ads channel contribute to this overall impact. Using channel spend or impression data at the campaign level, the sub-model estimates how much of the total contribution can be attributed to each level.\u003C\u002Fp>\n\n\n\n\u003Cp>This sub-model can remain relatively simple because it doesn’t need to account for factors like seasonality, macroeconomics, or other externalities, as these have already been handled in the initial channel-level MMM. Instead, the sub-model focuses on breaking down the internal mechanics of the channel itself, isolating the effects of specific campaigns or funnel stages. For example, if Google ads account for 10% of sales, the sub-model might reveal that \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fwhat-are-performance-max-campaigns\u002F\">performance max campaigns\u003C\u002Fa> contribute 7% of that, while \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fanatomy-of-a-google-display-campaign-and-how-to-make-it-work-harder\u002F\">display\u003C\u002Fa> and video campaigns contribute the other 3%. This layered approach allows for granular insights without reintroducing the external complexities that have already been modelled.\u003C\u002Fp>\n\n\n\n\u003Cp>The benefit of this method is that it allows for more granular insights without making the overall model overly complicated. It also leverages the broader MMM framework, ensuring that important external factors have been considered in the channel-level model before diving into more granular details. However, one limitation is that for a small number of campaign types, the sub-model could struggle to explain the contributions passed down from the initial model, as there may not be enough variation in the data to accurately distribute the impact across different campaigns. Additionally, since the sub-model is built on assumptions made in the initial MMM, any inaccuracies in the broader model could cascade down to the sub-level analysis.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"which-approach-should-you-take\">Which approach should you take?\u003C\u002Fh2>\n\n\n\n\u003Cp>The approach you should take depends on the resources and time you have available. If you have the time and commitment to conduct frequent, well-controlled experiments, the incrementality testing approach is generally best practice. It provides the most accurate insights into the specific contributions of channels and campaigns by directly measuring their causal impact. However, if frequent testing isn’t feasible due to time constraints, budget limitations, or the complexity of managing these experiments, sub-modelling is a more practical alternative. By leveraging existing spend or impression data to estimate contributions within the channel, sub-modelling allows you to gain granular insights while avoiding the need for ongoing experimentation. While incrementality testing is ideal for precision, sub-modelling offers a more scalable solution when resources are limited.\u003C\u002Fp>\n\n\n\n\u003Cp>In conclusion, expanding the granularity of your Marketing Mix Model beyond channel-level insights to capture campaign or funnel-specific contributions is essential for optimising marketing performance. Whether through incrementality testing or sub-modelling, both approaches offer valuable ways to dig deeper into your data without overwhelming your model with complexity. The key is to choose the method that best aligns with your \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing\u002F#when-should-you-consider-incrementality-testing\">resources and objectives\u003C\u002Fa>. By taking the right approach, you can enhance the effectiveness of your marketing strategy and make more informed decisions.\u003C\u002Fp>\n",{"rendered":151,"protected":211},30620,{"_yoast_wpseo_focuskw":5,"_yoast_wpseo_title":920,"_yoast_wpseo_metadesc":921,"inline_featured_image":211,"footnotes":5},"Balancing granularity in Marketing Mix Modelling | Impression","Discover the role of granularity in Marketing Mix Modelling, improving analysis precision and helping optimise budget allocation and marketing performance.",[730,220],[],[223],[926,113,226,227,228,229,230,735,232],"post-30611",[],[],[930,931,932,933],{"path":185,"name":186},{"path":238,"name":239},{"path":147,"name":148},{"path":194,"name":150},{"thumbnail":935,"medium":936,"medium_large":938,"large":940,"1536x1536":942,"2048x2048":944,"post-thumbnail":946,"12_col":948,"default_landscape":950,"default_portrait":952,"6_col_and_margin_short":954,"11_col_and_margin":956,"9_col_rectangle":958,"11_col":960,"staff_image_696_600":962,"rectangular_thumbnail":964},{"url":155,"width":244,"height":244},{"url":937,"width":70,"height":70},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=200&q=90&w=133",{"url":939,"width":73,"height":23},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1152&q=90&w=768",{"url":941,"width":251,"height":252},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1000&q=90&w=1892",{"url":943,"width":79,"height":79},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1536&q=90&w=1024",{"url":945,"width":82,"height":82},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=2048&q=90&w=1365",{"url":947,"width":85,"height":86},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1180&q=90&rect=0%2C1777%2C2400%2C1522&w=1860",{"url":949,"width":88,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2880",{"url":951,"width":91,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=800",{"url":953,"width":65,"height":91},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=800&q=90&w=600",{"url":955,"width":85,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=1860",{"url":957,"width":95,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=3080",{"url":959,"width":97,"height":98},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1400&q=90&w=2150",{"url":961,"width":100,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2640",{"url":963,"width":64,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=696",{"url":965,"width":103,"height":104},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=300&q=90&rect=0%2C1910%2C2400%2C960&w=750",[967],{"id":213,"path":279,"display_name":4,"job_title":11,"description":6,"avatar_url":122,"acf":968},{"full_bio":10,"job_title":11,"specialist_service_links":969,"photo":976},[970,972,974],{"page":971,"link_text":29},{"ID":15,"post_author":16,"post_date":17,"post_date_gmt":17,"post_content":5,"post_title":18,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":21,"to_ping":5,"pinged":5,"post_modified":22,"post_modified_gmt":22,"post_content_filtered":5,"post_parent":23,"guid":24,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":28},{"page":973,"link_text":34},{"ID":32,"post_author":16,"post_date":33,"post_date_gmt":33,"post_content":5,"post_title":34,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":35,"to_ping":5,"pinged":5,"post_modified":36,"post_modified_gmt":36,"post_content_filtered":5,"post_parent":37,"guid":38,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":39},{"page":975,"link_text":44},{"ID":42,"post_author":16,"post_date":43,"post_date_gmt":43,"post_content":5,"post_title":44,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":45,"to_ping":5,"pinged":5,"post_modified":46,"post_modified_gmt":46,"post_content_filtered":5,"post_parent":37,"guid":47,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":48},{"ID":50,"id":50,"title":51,"filename":52,"filesize":53,"url":54,"link":55,"alt":5,"author":56,"description":5,"caption":5,"name":57,"status":58,"uploaded_to":23,"date":59,"modified":59,"menu_order":23,"mime_type":60,"type":61,"subtype":62,"icon":63,"width":64,"height":65,"sizes":977},{"thumbnail":67,"thumbnail-width":68,"thumbnail-height":68,"medium":69,"medium-width":70,"medium-height":71,"medium_large":72,"medium_large-width":73,"medium_large-height":74,"large":75,"large-width":76,"large-height":77,"1536x1536":78,"1536x1536-width":79,"1536x1536-height":80,"2048x2048":81,"2048x2048-width":82,"2048x2048-height":83,"post-thumbnail":84,"post-thumbnail-width":85,"post-thumbnail-height":86,"12_col":87,"12_col-width":88,"12_col-height":89,"default_landscape":90,"default_landscape-width":91,"default_landscape-height":65,"default_portrait":92,"default_portrait-width":65,"default_portrait-height":91,"6_col_and_margin_short":93,"6_col_and_margin_short-width":85,"6_col_and_margin_short-height":89,"11_col_and_margin":94,"11_col_and_margin-width":95,"11_col_and_margin-height":89,"9_col_rectangle":96,"9_col_rectangle-width":97,"9_col_rectangle-height":98,"11_col":99,"11_col-width":100,"11_col-height":89,"staff_image_696_600":101,"staff_image_696_600-width":64,"staff_image_696_600-height":65,"rectangular_thumbnail":102,"rectangular_thumbnail-width":103,"rectangular_thumbnail-height":104},[979,980],{"id":730,"name":148,"path":147},{"id":220,"name":136,"path":135},[982,983,984,985,987,988,989,990,991,992,993,994,996,998,1000,1001,1002,1003,1004,1005,1006,1007],{"property":296,"content":920},{"property":298,"content":921},{"name":300,"content":301},{"property":199,"content":986},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgranularity-marketing-mix-modelling\u002F",{"name":305,"content":306},{"property":308,"content":309},{"property":311,"content":920},{"property":313,"content":921},{"property":315,"content":986},{"property":317,"content":318},{"property":320,"content":321},{"property":323,"content":995},"2024-10-03T16:37:10+00:00",{"property":326,"content":997},"2024-10-03T16:37:11+00:00",{"property":329,"content":999},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fopen-graph-granularity-in-mmm.png?auto=compress%2Cformat&q=90",{"property":332,"content":333},{"property":335,"content":336},{"property":338,"content":60},{"property":180,"content":4},{"property":341,"content":342},{"property":344,"content":999},{"property":346,"content":347},{"property":349,"content":347},{"@context":351,"@graph":1009},[1010,1021,1027,1029,1035,1039,1057],{"@type":290,"@id":1011,"isPartOf":1012,"author":1013,"headline":150,"datePublished":995,"dateModified":997,"mainEntityOfPage":1014,"wordCount":1015,"publisher":1016,"image":1017,"thumbnailUrl":1019,"articleSection":1020,"inLanguage":366},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgranularity-marketing-mix-modelling\u002F#article",{"@id":986},{"name":4,"@id":357},{"@id":986},1673,{"@id":361},{"@id":1018},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgranularity-marketing-mix-modelling\u002F#primaryimage","https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F10\u002Fnathan-dumlao-PHl9ObJQo38-unsplash.jpg?auto=compress%2Cformat&q=90",[148,136],{"@type":368,"@id":986,"url":986,"name":920,"isPartOf":1022,"primaryImageOfPage":1023,"image":1024,"thumbnailUrl":1019,"datePublished":995,"dateModified":997,"description":921,"breadcrumb":1025,"inLanguage":375},{"@id":370},{"@id":1018},{"@id":1018},{"@id":1026},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgranularity-marketing-mix-modelling\u002F#breadcrumb",{"@type":377,"inLanguage":366,"@id":1018,"url":1019,"contentUrl":1019,"width":378,"height":1028},3600,{"@type":381,"@id":1026,"itemListElement":1030},[1031,1032,1033,1034],{"@type":384,"position":385,"name":186,"item":386},{"@type":384,"position":388,"name":239,"item":389},{"@type":384,"position":391,"name":148,"item":843},{"@type":384,"position":394,"name":150},{"@type":396,"@id":370,"url":386,"name":318,"description":5,"publisher":1036,"inLanguage":366,"potentialAction":1037},{"@id":398},{"@type":400,"target":1038,"query-input":404},{"@type":402,"urlTemplate":403},{"@type":406,"@id":398,"name":318,"url":386,"logo":1040,"image":412,"sameAs":1041,"@context":417,"email":418,"foundingDate":419,"legalName":420,"vatID":421,"description":422,"telephone":423,"address":1042,"location":1043,"areaServed":1053},{"@type":377,"inLanguage":366,"@id":408,"url":409,"contentUrl":409,"width":410,"height":411,"caption":318},[414,415,416],{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},[1044,1047,1050],{"@type":433,"name":434,"geo":1045,"address":1046,"photo":412,"telephone":440},{"@type":436,"latitude":437,"longitude":438},{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},{"@type":433,"name":442,"geo":1048,"address":1049,"photo":451,"telephone":452},{"@type":436,"latitude":444,"longitude":445},{"@type":425,"streetAddress":447,"addressLocality":448,"addressRegion":449,"postalCode":450,"addressCountry":430},{"@type":433,"name":454,"geo":1051,"address":1052},{"@type":436,"latitude":456,"longitude":457},{"@type":425,"streetAddress":459,"addressLocality":460,"addressRegion":461,"postalCode":462,"addressCountry":463},[1054,1055,1056],{"@type":466,"name":467},{"@type":466,"name":469},{"@type":466,"name":463},{"@type":472,"@id":473,"name":4,"image":1058,"description":6,"sameAs":1059,"url":200},{"@type":377,"inLanguage":366,"@id":475,"url":8,"contentUrl":8,"caption":4},[196],{"self":1061,"collection":1066,"about":1068,"author":1070,"replies":1072,"version-history":1075,"predecessor-version":1078,"wp:featuredmedia":1082,"wp:attachment":1085,"wp:term":1088,"curies":1095},[1062],{"href":1063,"targetHints":1064},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F30611",{"allow":1065},[483,484,485,486,487],[1067],{"href":490},[1069],{"href":493},[1071],{"embeddable":108,"href":496},[1073],{"embeddable":108,"href":1074},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcomments?post=30611",[1076],{"count":173,"href":1077},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F30611\u002Frevisions",[1079],{"id":1080,"href":1081},30619,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F30611\u002Frevisions\u002F30619",[1083],{"embeddable":108,"href":1084},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia\u002F30620",[1086],{"href":1087},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia?parent=30611",[1089,1091,1093],{"taxonomy":516,"embeddable":108,"href":1090},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcategories?post=30611",{"taxonomy":519,"embeddable":108,"href":1092},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcollections?post=30611",{"taxonomy":180,"embeddable":108,"href":1094},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcoauthors?post=30611",[1096],{"name":525,"href":526,"templated":108},{"id":1098,"date":1099,"date_gmt":1099,"guid":1100,"modified":163,"modified_gmt":163,"slug":1102,"status":19,"type":113,"link":162,"title":1103,"content":1104,"excerpt":1106,"author":213,"featured_media":1107,"comment_status":20,"ping_status":20,"sticky":211,"template":5,"format":215,"meta":1108,"categories":1111,"collections":1112,"coauthors":1113,"class_list":1114,"acf":1116,"head_links":1117,"breadcrumbs":1118,"url_path":162,"search_description":1110,"featured_image_urls":1123,"authors":1155,"type_nicename":290,"reading_minutes":142,"category":1167,"yoast_title":1109,"yoast_meta":1170,"yoast_json_ld":1197,"_links":1249},29789,"2024-08-01T14:45:33",{"rendered":1101},"\u002F?p=29789","google-big-query-ml",{"rendered":160},{"rendered":1105,"protected":211},"\n\u003Cp>\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fcloud.google.com\u002Fbigquery\u002Fdocs\u002Fbqml-introduction\">Google BigQuery ML\u003C\u002Fa> (BQML) is a powerful tool that allows data scientists and analysts to build and deploy machine learning models directly within Google BigQuery. By integrating machine learning into BigQuery, BQML eliminates the need to move data between various systems or platforms, enabling you to utilise \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Fdata-science\u002F\">data science\u003C\u002Fa> to extract data-driven insights efficiently. This simplifies the workflow and reduces the time to develop and operationalise ML models.\u003C\u002Fp>\n\n\n\n\u003Cp>You can build and train models using \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FSQL\">SQL\u003C\u002Fa> queries, allowing for easy integration with existing \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Fmodern-data-stack\u002F\">data pipelines\u003C\u002Fa>. BQML supports a variety of model types, including linear and logistic regression, ARIMA models, and more, enabling you to tackle a wide range of predictive analytics tasks efficiently.\u003C\u002Fp>\n\n\n\u003Cdiv class=\"wp-block-image\">\n\u003Cfigure class=\"aligncenter size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1239\" height=\"1075\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fbig-query-ml-1.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-29817\"\u002F>\u003C\u002Ffigure>\u003C\u002Fdiv>\n\n\n\u003Cp>Using BigQuery ML for data analysis is made significantly easier due to its seamless integration with Google \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fanalytics\u002Flooker-studio\u002F\">Data Studio\u003C\u002Fa>, Google’s powerful data visualisation and business intelligence tool. This connection allows users to directly visualise, explore, and share insights from their BigQuery ML models without the need for complex data transfers or additional processing. Data Studio provides an intuitive, drag-and-drop interface for creating rich, interactive dashboards that can display real-time results from BQML models.\u003C\u002Fp>\n\n\n\n\u003Cp>Additionally, you can use BigQuery ML to provide some elementary insights into \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Fmeasurement\u002F\">media effectiveness\u003C\u002Fa> by leveraging its machine learning capabilities to analyse and optimise media strategies. It enables quick predictive modelling to forecast campaign impacts, audience segmentation for tailored targeting, and \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fhow-data-driven-attribution-works-in-google-analytics-4\u002F\">multi-touch attribution\u003C\u002Fa> to understand the contribution of each media touchpoint. Its seamless integration with tools like Data Studio and \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fgoogle-analytics-4-updates-and-what-they-mean-for-your-business\u002F\">Google Analytics\u003C\u002Fa> ensures comprehensive analysis and easy implementation of insights.\u003C\u002Fp>\n\n\n\n\u003Cdiv class=\"table-of-contents\">\u003Cul>\u003Cli class=\"heading-h2\">\u003Ca href=\"#key-functions\">Key functions\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#the-create-model-statement\">The “CREATE MODEL” statement?\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#the-ml-evaluate-function\">The “ML.EVALUATE” function\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#the-ml-predict-function\">The “ML.PREDICT” function\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#time-series-forecasting\">Time series forecasting\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#introduction-and-objectives\">Introduction and objectives\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#model-creation-and-evaluation\">Model creation and evaluation\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#forecasting\">Forecasting\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#logic\">Logic regression\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#introduction-and-objectives\">Introduction and objectives\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#model-creation-and-evaluation\">Model creation and evaluation\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h3\">\u003Ca href=\"#prediction\">Prediction\u003C\u002Fa>\u003C\u002Fli>\u003Cli class=\"heading-h2\">\u003Ca href=\"#conclusion\">Conclusion\u003C\u002Fa>\u003C\u002Fli>\u003C\u002Ful>\u003C\u002Fdiv>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"key-functions\">Key functions\u003C\u002Fh2>\n\n\n\n\u003Cp>Google BQML breaks the task of building and operationalising ML models into 3 steps, each associated with its own function. The typical workflow in BQML begins with the creation of the model, it is here that parameter estimation takes place. This is then followed by an \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fbuilding-trustworthy-media-insights-through-model-validation\u002F\">evaluation of the model\u003C\u002Fa>, in which precision and accuracy are analysed. Lastly, the model is then used to perform prediction. \u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"the-create-model-statement\">The “CREATE MODEL” statement \u003C\u002Fh3>\n\n\n\n\u003Cp>The first step when developing any ML model is to build it, which involves parameter estimation. During the creation of the model, BQML automatically splits the data into \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FTraining,_validation,_and_test_data_sets\">training and testing subsets\u003C\u002Fa>, enabling you to assess the predictive accuracy of your given model. Parameter estimation is performed in BQML by minimising the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.statistics.com\u002Fglossary\u002Floss-function\u002F#:~:text=A%20loss%20function%20specifies%20a,accurate%20within%20a%20certain%20range.\">loss statistic\u003C\u002Fa>, a metric describing how well the model predicts the test dataset. If predictions are close to the unseen data then the loss should be close to 0.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"902\" height=\"678\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fbig-query-ml-2.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-29818\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Cp>Additionally, the “\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fcloud.google.com\u002Fbigquery\u002Fdocs\u002Freference\u002Fstandard-sql\u002Fbigqueryml-syntax-create\">CREATE MODEL\u003C\u002Fa>” statement is where you specify the type of model you wish to build. The model you build should be relevant to the objective in mind. BQML has provided \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fcloud.google.com\u002Fbigquery\u002Fdocs\u002Fbqml-introduction#model_selection_guide\">this guide\u003C\u002Fa> to help with selecting the right model.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"the-ml-evaluate-function\">The “ML.EVALUATE” function\u003C\u002Fh3>\n\n\n\n\u003Cp>The second step in the ML workflow is to evaluate the performance of the model, which is done in BQML using the “\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fcloud.google.com\u002Fbigquery\u002Fdocs\u002Freference\u002Fstandard-sql\u002Fbigqueryml-syntax-evaluate\">ML.EVALUATE\u003C\u002Fa>” function. This function evaluates the predicted values generated by the model against the actual data for both the training and the testing datasets. Outputs may vary depending on the type of model you build, the following metrics are standard output when working with a \u003Ca rel=\"noreferrer noopener\" href=\"#logic\">logistic regression model\u003C\u002Fa>, in which binary classification is the objective in mind:\u003Cbr>\u003C\u002Fp>\n\n\n\n\u003Cul class=\"wp-block-list\">\n\u003Cli>\u003Cstrong>precision: \u003C\u002Fstrong>a metric for classification models which identifies the frequency with which a model was correct when predicting the positive class.\u003C\u002Fli>\n\n\n\n\u003Cli>\u003Cstrong>recall: \u003C\u002Fstrong>a metric for classification models that tells you how many positive labels, out of all the possible positive labels, you correctly identified. \u003C\u002Fli>\n\n\n\n\u003Cli>\u003Cstrong>accuracy: \u003C\u002Fstrong>a metric for classification models that tells you the fraction of predictions that a classification model got right.\u003C\u002Fli>\n\n\n\n\u003Cli>\u003Cstrong>f1_score: \u003C\u002Fstrong>measures the overall accuracy of the model. An f1 score lies between 0 and 1, 1 indicating optimal accuracy.\u003C\u002Fli>\n\n\n\n\u003Cli>\u003Cstrong>log_loss:\u003C\u002Fstrong> This is the measure of how far the model’s predictions are from the correct labels. The log is used due to the nature of logistic regression.\u003C\u002Fli>\n\n\n\n\u003Cli>\u003Cstrong>roc_auc: \u003C\u002Fstrong>refers to the area under the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fdevelopers.google.com\u002Fmachine-learning\u002Fcrash-course\u002Fclassification\u002Froc-and-auc#:~:text=An%20ROC%20curve%20(receiver%20operating,False%20Positive%20Rate\">ROC curve\u003C\u002Fa>, which is the probability that a classifier is more confident that a randomly chosen positive example is actually positive than that a randomly chosen negative example is positive.\u003C\u002Fli>\n\u003C\u002Ful>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"the-ml-predict-function\">The “ML.PREDICT” function\u003C\u002Fh3>\n\n\n\n\u003Cp>The “\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fcloud.google.com\u002Fbigquery\u002Fdocs\u002Freference\u002Fstandard-sql\u002Fbigqueryml-syntax-predict\">ML.PREDICT\u003C\u002Fa>” function allows you to forecast from the model, however, care must be taken with forecasting as ML models tend to lose predictive accuracy the further away we \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FExtrapolation\">extrapolate\u003C\u002Fa> from the known data set. By applying this function, you can easily predict outcomes based on new input data directly within BigQuery.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"time-series-forecasting\">Time series forecasting\u003C\u002Fh2>\n\n\n\n\u003Cp>Time series forecasting is crucial in digital marketing as it enables you to predict future trends and customer behaviours based on historical data, allowing for more informed decision-making and strategic \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fdigital-marketing\u002Fstrategy\u002Fmedia-planning\u002F\">media planning\u003C\u002Fa>. By leveraging time series forecasting, you can anticipate peak periods, optimise campaign timing, and allocate budgets more effectively to maximise ROAS.\u003C\u002Fp>\n\n\n\n\u003Cp>Integrating time series forecasting with \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fmedia-mix-modelling\u002F\">Marketing Mix Modeling\u003C\u002Fa> (MMM) further enhances its value, as MMM helps quantify the impact of various marketing channels and activities on sales and other key performance indicators. This combination allows you to not only forecast future outcomes but also understand the drivers behind these predictions, enabling a more holistic and data-driven approach to campaign optimization and resource allocation. \u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"introduction-and-objectives\">Introduction and objectives\u003C\u002Fh3>\n\n\n\n\u003Cp>In this example, we’ll create an \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FAutoregressive_integrated_moving_average\">ARIMA\u003C\u002Fa> time series model to perform single time-series forecasting using the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fconsole.cloud.google.com\u002Fbigquery?p=bigquery-public-data&d=google_analytics_sample&t=ga_sessions_20170801&page=table&_ga=2.128198468.936982125.1721810777-961047795.1695717430&_gac=1.128132734.1720107963.CjwKCAjwkJm0BhBxEiwAwT1AXNKlycW5alGJF6nTcAc_vuj3xMq6gyj_6msF9NthFDiThCCRFaERyxoCPIoQAvD_BwE\">google_analytics_sample.ga_sessions\u003C\u002Fa> sample table. The ga_sessions table contains information about a slice of session data collected by \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fanalytics\u002Fgoogle-analytics-360\u002F\">Google Analytics 360\u003C\u002Fa> and sent to BigQuery. We’ve extracted the daily number of website visits from 2016-08-01 to 2017-08-01. \u003C\u002Fp>\n\n\n\n\u003Cp>The objective is to forecast the number of website visits for August 2017.\u003C\u002Fp>\n\n\n\n\u003Cp>The first step in creating any form of ML model is to understand the data you’re modelling. The visual below shows the data we’ll be modelling.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"933\" height=\"412\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fbig-query-ml-3.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-29819\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Cp>At first glance, this time series exhibits an increasing trend until ~November 2016. This trend then starts to decrease until December 2016, telling us that we have a non-constant mean. This indicates that perhaps taking the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fotexts.com\u002Ffpp2\u002Fstationarity.html#differencing\">first difference\u003C\u002Fa> might eliminate the trends present in this data. \u003C\u002Fp>\n\n\n\n\u003Cp>Furthermore, the variance of this time series does appear to be somewhat constant over time, with very few peaks and troughs standing out. This indicates no need for a transform to eliminate fluctuations in variance. \u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"model-creation-and-evaluation\">Model creation and evaluation\u003C\u002Fh3>\n\n\n\n\u003Cp>Now we’ve familiarised ourselves with the data and identified the type of model we require, the next step is to run the “CREATE model” statement. As we’re going to use an ARIMA model we’ll need to specify “model_type = ‘ARIMA_PLUS’”. By default, the algorithm fits dozens of candidate models and automatically tunes the hyperparameters. BQML chooses the model with the lowest \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FAkaike_information_criterion\">Akaike information criterion (AIC)\u003C\u002Fa>.\u003C\u002Fp>\n\n\n\n\u003Cp>After creating your model, the next step is to evaluate its performance. To do this we simply run the “ML.EVALUATE” function and observe our outputs. Upon running this, you should find the outputs of up to 42 candidate models, notice that the optimal model is the one with the lowest AIC which is always on row 1. The relevant outputs for row 1 are as follows:\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-table\">\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>p\u003C\u002Fth>\u003Cth>d\u003C\u002Fth>\u003Cth>q\u003C\u002Fth>\u003Cth>\u003Cstrong>has_drift\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>AIC\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>variance\u003C\u002Fstrong>\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>2\u003C\u002Ftd>\u003Ctd>1\u003C\u002Ftd>\u003Ctd>3\u003C\u002Ftd>\u003Ctd>true\u003C\u002Ftd>\u003Ctd>4885.81\u003C\u002Ftd>\u003Ctd>36334\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Ffigure>\n\n\n\n\u003Cp>This tells us we’re forecasting from an \u003Cem>ARIMA\u003C\u002Fem>(2,1,3) with an intercept term. \u003C\u002Fp>\n\n\n\n\u003Cp>For more information on the nature of our model, we can run the “ML.ARIMA_COEFFICIENTS” function to extract the values of the coefficients of our model as tabulated below.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-table\">\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>\u003Cstrong>ma1\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>ma2\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>ar1\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>ar2\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>ar3\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>intercept\u003C\u002Fstrong>\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>-0.1498\u003C\u002Ftd>\u003Ctd>-0.6834\u003C\u002Ftd>\u003Ctd>-0.2523\u003C\u002Ftd>\u003Ctd>-0.6894\u003C\u002Ftd>\u003Ctd>-0.6484\u003C\u002Ftd>\u003Ctd>2.6125\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Ffigure>\n\n\n\n\u003Cp>Overall, our model has 2 non-seasonal \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FMoving-average_model\">moving average components\u003C\u002Fa>, 3 non-seasonal \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FAutoregressive_model\">autoregressive components\u003C\u002Fa> and has non-seasonal difference of order 1. This model is optimal, suggesting that there is no need for seasonal components for forecasting.\u003C\u002Fp>\n\n\n\n\u003Cp>It is important to point out here that when building an ML model, model diagnostics should be a part of your workflow! Fortunately for us, the “ML.EXPLAIN.FORECAST” function allows you to extract your residuals for conducting \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.geeksforgeeks.org\u002Fresidual-analysis\u002F\">residual analysis\u003C\u002Fa>.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"forecasting\">Forecasting\u003C\u002Fh3>\n\n\n\n\u003Cp>Now we have evaluated our model and are familiar with its various components, then the next step is to forecast the number of website visits for August 2017. To do this we make use of the “ML.FORECAST” function, which produces both future time series values and prediction intervals. The following forecasts were produced.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"937\" height=\"415\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fbig-query-ml-4.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-29820\"\u002F>\u003C\u002Ffigure>\n\n\n\n\u003Cp>Now that we have our daily forecasts, we may wish to summarise them for the whole of August. The table below shows the forecasted total number of website visits, along with the associated upper and lower bounds. \u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-table\">\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth scope=\"col\">Total mean forecast\u003C\u002Fth>\u003Cth scope=\"col\">Total upper forecast\u003C\u002Fth>\u003Cth scope=\"col\">Total lower forecast\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>~72,029\u003C\u002Ftd>\u003Ctd>~57,667\u003C\u002Ftd>\u003Ctd>~86,392\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Ffigure>\n\n\n\n\u003Cp>On average we expect to see 72,029 website visits during August 2017. We should expect to see at least 57,667 and at most 86,392.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"logic\">Logic regression\u003C\u002Fh2>\n\n\n\n\u003Cp>Logistic regression is a particularly valuable tool to have in digital marketing as it provides a robust method for predicting binary outcomes, such as \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fdigital-experience\u002Fcro\u002F\">customer conversion\u003C\u002Fa> or click-through rates, based on various predictor variables like user behaviour, demographic information, and past interactions. Through logistic regression, you can identify the key factors that influence customer decisions and tailor their strategies to target high-probability converters more effectively. \u003C\u002Fp>\n\n\n\n\u003Cp>Additionally, logistic regression can be seamlessly integrated with \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fdemystifying-rfm-segmentation-analysis\u002F\">RFM\u003C\u002Fa> (Recency, Frequency, Monetary) modelling to predict customer behaviours such as purchase likelihood, enabling you to target and engage the most valuable customer segments more effectively.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"introduction-and-objectives\">Introduction and objectives\u003C\u002Fh3>\n\n\n\n\u003Cp>In this example we create a logistic regression model using BigQuery ML. We’ll use the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fsupport.google.com\u002Fanalytics\u002Fanswer\u002F7586738?ref_topic=3416089#zippy=%2Cin-this-article\">Google Analytics sample dataset\u003C\u002Fa> to create a model that predicts whether a website visitor will make a transaction. If the number of transactions within the session is “NULL”, we assign this to a value of 0. Otherwise, it is set to 1. These values represent the possible outcomes.\u003C\u002Fp>\n\n\n\n\u003Cp>Given we only have existing data up to 30th June 2017, the objective of this model is to predict the number of website visitors that will make a transaction throughout July 2017.\u003C\u002Fp>\n\n\n\n\u003Cp>To model the probability that a website visitor will complete a transaction, the following predictor variables have been used:\u003C\u002Fp>\n\n\n\n\u003Cul class=\"wp-block-list\">\n\u003Cli>The operating system of the visitor’s device.\u003C\u002Fli>\n\n\n\n\u003Cli>The device they used to visit the website.\u003C\u002Fli>\n\n\n\n\u003Cli>The country from which the session originated.\u003C\u002Fli>\n\n\n\n\u003Cli>The number of page views within that session.\u003C\u002Fli>\n\u003C\u002Ful>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"model-creation-and-evaluation\">Model creation and evaluation\u003C\u002Fh3>\n\n\n\n\u003Cp>Now we have all our variables defined and have identified the type of model we require, the next step is to run the “CREATE model” statement. We’re running the model using daily data from 2016-08-01 to 2017-06-30.\u003C\u002Fp>\n\n\n\n\u003Cp>After creating your model the next step is to evaluate its performance. To do this we simply run the “ML.EVALUATE” function and observe our outputs. The outputs of doing so are tabulated below:\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-table\">\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>\u003Cstrong>precision\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>recall\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>accuracy\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>f1_score\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>log_loss\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>roc_auc\u003C\u002Fstrong>\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>0.4685\u003C\u002Ftd>\u003Ctd>0.1108\u003C\u002Ftd>\u003Ctd>0.9853\u003C\u002Ftd>\u003Ctd>0.1792\u003C\u002Ftd>\u003Ctd>0.0462\u003C\u002Ftd>\u003Ctd>0.9817\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Ffigure>\n\n\n\n\u003Cp>As we can see, the model metrics here are quite low and significantly far away from the maximums, indicating we have plenty of room for improvement and should reconsider. For a “good” model, we should expect most of these metrics (apart from the log_loss) to be close to 1. For the purposes of this blog, we’re going to continue with this model and produce some predictions.\u003C\u002Fp>\n\n\n\n\u003Ch3 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"prediction\">Prediction\u003C\u002Fh3>\n\n\n\n\u003Cp>Using the “ML.PREDICT” function, we’ve predicted the number of transactions made by website visitors from the USA for July 2017.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-table\">\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>\u003Cstrong>Country\u003C\u002Fstrong>\u003C\u002Fth>\u003Cth>\u003Cstrong>Number of Transactions\u003C\u002Fstrong>\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>USA\u003C\u002Ftd>\u003Ctd>220\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Ffigure>\n\n\n\n\u003Cp>Therefore the number of transactions made by website visitors from the USA for July 2017 is predicted to be approximately 220.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" class=\"wp-block-heading\" id=\"conclusion\">Conclusion\u003C\u002Fh2>\n\n\n\n\u003Cp>In conclusion, Google BigQuery ML offers a robust easy-to-use platform for integrating machine learning directly within your data warehouse environment. By leveraging the CREATE MODEL, ML.EVALUATE, and ML.PREDICT functions, you can seamlessly build, evaluate, and deploy models using standard SQL queries. The example of ARIMA forecasting demonstrates the powerful capabilities of BQML for time series analysis, allowing for accurate forecasts and insightful analytics. \u003C\u002Fp>\n\n\n\n\u003Cp>Whether you are looking to perform classification, regression, or time series forecasting, BigQuery ML provides the tools necessary to streamline your machine learning workflows, making advanced analytics accessible and efficient.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Fmeasurement\u002F\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"3840\" height=\"540\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F04\u002Ff0sKYLsk-Measurement_Banner_01.jpg?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-27925\"\u002F>\u003C\u002Fa>\u003C\u002Ffigure>\n",{"rendered":161,"protected":211},29830,{"_yoast_wpseo_focuskw":5,"_yoast_wpseo_title":1109,"_yoast_wpseo_metadesc":1110,"inline_featured_image":211,"footnotes":5},"A deep dive into Google Big Query ML | Impression","Google BigQuery ML is a powerful tool that allows data scientists and analysts to build and deploy machine learning models directly within Google BigQuery.",[730,220],[],[223],[1115,113,226,227,228,229,230,735,232],"post-29789",[],[],[1119,1120,1121,1122],{"path":185,"name":186},{"path":238,"name":239},{"path":135,"name":136},{"path":194,"name":160},{"thumbnail":1124,"medium":1125,"medium_large":1127,"large":1129,"1536x1536":1131,"2048x2048":1133,"post-thumbnail":1135,"12_col":1137,"default_landscape":1139,"default_portrait":1141,"6_col_and_margin_short":1143,"11_col_and_margin":1145,"9_col_rectangle":1147,"11_col":1149,"staff_image_696_600":1151,"rectangular_thumbnail":1153},{"url":164,"width":244,"height":244},{"url":1126,"width":70,"height":70},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=150&q=90&w=200",{"url":1128,"width":73,"height":23},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=576&q=90&w=768",{"url":1130,"width":251,"height":252},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1000&q=90&w=1892",{"url":1132,"width":79,"height":79},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1152&q=90&w=1536",{"url":1134,"width":82,"height":82},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1536&q=90&w=2048",{"url":1136,"width":85,"height":86},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1180&q=90&w=1860",{"url":1138,"width":88,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2880",{"url":1140,"width":91,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=800",{"url":1142,"width":65,"height":91},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=800&q=90&w=600",{"url":1144,"width":85,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=1860",{"url":1146,"width":95,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=3080",{"url":1148,"width":97,"height":98},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1400&q=90&w=2150",{"url":1150,"width":100,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2640",{"url":1152,"width":64,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=696",{"url":1154,"width":103,"height":104},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=300&q=90&w=750",[1156],{"id":213,"path":279,"display_name":4,"job_title":11,"description":6,"avatar_url":122,"acf":1157},{"full_bio":10,"job_title":11,"specialist_service_links":1158,"photo":1165},[1159,1161,1163],{"page":1160,"link_text":29},{"ID":15,"post_author":16,"post_date":17,"post_date_gmt":17,"post_content":5,"post_title":18,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":21,"to_ping":5,"pinged":5,"post_modified":22,"post_modified_gmt":22,"post_content_filtered":5,"post_parent":23,"guid":24,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":28},{"page":1162,"link_text":34},{"ID":32,"post_author":16,"post_date":33,"post_date_gmt":33,"post_content":5,"post_title":34,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":35,"to_ping":5,"pinged":5,"post_modified":36,"post_modified_gmt":36,"post_content_filtered":5,"post_parent":37,"guid":38,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":39},{"page":1164,"link_text":44},{"ID":42,"post_author":16,"post_date":43,"post_date_gmt":43,"post_content":5,"post_title":44,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":45,"to_ping":5,"pinged":5,"post_modified":46,"post_modified_gmt":46,"post_content_filtered":5,"post_parent":37,"guid":47,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":48},{"ID":50,"id":50,"title":51,"filename":52,"filesize":53,"url":54,"link":55,"alt":5,"author":56,"description":5,"caption":5,"name":57,"status":58,"uploaded_to":23,"date":59,"modified":59,"menu_order":23,"mime_type":60,"type":61,"subtype":62,"icon":63,"width":64,"height":65,"sizes":1166},{"thumbnail":67,"thumbnail-width":68,"thumbnail-height":68,"medium":69,"medium-width":70,"medium-height":71,"medium_large":72,"medium_large-width":73,"medium_large-height":74,"large":75,"large-width":76,"large-height":77,"1536x1536":78,"1536x1536-width":79,"1536x1536-height":80,"2048x2048":81,"2048x2048-width":82,"2048x2048-height":83,"post-thumbnail":84,"post-thumbnail-width":85,"post-thumbnail-height":86,"12_col":87,"12_col-width":88,"12_col-height":89,"default_landscape":90,"default_landscape-width":91,"default_landscape-height":65,"default_portrait":92,"default_portrait-width":65,"default_portrait-height":91,"6_col_and_margin_short":93,"6_col_and_margin_short-width":85,"6_col_and_margin_short-height":89,"11_col_and_margin":94,"11_col_and_margin-width":95,"11_col_and_margin-height":89,"9_col_rectangle":96,"9_col_rectangle-width":97,"9_col_rectangle-height":98,"11_col":99,"11_col-width":100,"11_col-height":89,"staff_image_696_600":101,"staff_image_696_600-width":64,"staff_image_696_600-height":65,"rectangular_thumbnail":102,"rectangular_thumbnail-width":103,"rectangular_thumbnail-height":104},[1168,1169],{"id":730,"name":148,"path":147},{"id":220,"name":136,"path":135},[1171,1172,1173,1174,1176,1177,1178,1179,1180,1181,1182,1183,1185,1187,1189,1190,1191,1192,1193,1194,1195,1196],{"property":296,"content":1109},{"property":298,"content":1110},{"name":300,"content":301},{"property":199,"content":1175},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgoogle-big-query-ml\u002F",{"name":305,"content":306},{"property":308,"content":309},{"property":311,"content":1109},{"property":313,"content":1110},{"property":315,"content":1175},{"property":317,"content":318},{"property":320,"content":321},{"property":323,"content":1184},"2024-08-01T14:45:33+00:00",{"property":326,"content":1186},"2026-08-25T09:29:44+00:00",{"property":329,"content":1188},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fopen-graph-big-query-ml.png?auto=compress%2Cformat&q=90",{"property":332,"content":333},{"property":335,"content":336},{"property":338,"content":60},{"property":180,"content":4},{"property":341,"content":342},{"property":344,"content":1188},{"property":346,"content":347},{"property":349,"content":347},{"@context":351,"@graph":1198},[1199,1210,1216,1218,1224,1228,1246],{"@type":290,"@id":1200,"isPartOf":1201,"author":1202,"headline":160,"datePublished":1184,"dateModified":1186,"mainEntityOfPage":1203,"wordCount":1204,"publisher":1205,"image":1206,"thumbnailUrl":1208,"articleSection":1209,"inLanguage":366},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgoogle-big-query-ml\u002F#article",{"@id":1175},{"name":4,"@id":357},{"@id":1175},1995,{"@id":361},{"@id":1207},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgoogle-big-query-ml\u002F#primaryimage","https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F08\u002Fadrien-delforge-yZrhORWUATg-unsplash.jpg?auto=compress%2Cformat&q=90",[148,136],{"@type":368,"@id":1175,"url":1175,"name":1109,"isPartOf":1211,"primaryImageOfPage":1212,"image":1213,"thumbnailUrl":1208,"datePublished":1184,"dateModified":1186,"description":1110,"breadcrumb":1214,"inLanguage":375},{"@id":370},{"@id":1207},{"@id":1207},{"@id":1215},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fgoogle-big-query-ml\u002F#breadcrumb",{"@type":377,"inLanguage":366,"@id":1207,"url":1208,"contentUrl":1208,"width":378,"height":1217},1800,{"@type":381,"@id":1215,"itemListElement":1219},[1220,1221,1222,1223],{"@type":384,"position":385,"name":186,"item":386},{"@type":384,"position":388,"name":239,"item":389},{"@type":384,"position":391,"name":136,"item":651},{"@type":384,"position":394,"name":160},{"@type":396,"@id":370,"url":386,"name":318,"description":5,"publisher":1225,"inLanguage":366,"potentialAction":1226},{"@id":398},{"@type":400,"target":1227,"query-input":404},{"@type":402,"urlTemplate":403},{"@type":406,"@id":398,"name":318,"url":386,"logo":1229,"image":412,"sameAs":1230,"@context":417,"email":418,"foundingDate":419,"legalName":420,"vatID":421,"description":422,"telephone":423,"address":1231,"location":1232,"areaServed":1242},{"@type":377,"inLanguage":366,"@id":408,"url":409,"contentUrl":409,"width":410,"height":411,"caption":318},[414,415,416],{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},[1233,1236,1239],{"@type":433,"name":434,"geo":1234,"address":1235,"photo":412,"telephone":440},{"@type":436,"latitude":437,"longitude":438},{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},{"@type":433,"name":442,"geo":1237,"address":1238,"photo":451,"telephone":452},{"@type":436,"latitude":444,"longitude":445},{"@type":425,"streetAddress":447,"addressLocality":448,"addressRegion":449,"postalCode":450,"addressCountry":430},{"@type":433,"name":454,"geo":1240,"address":1241},{"@type":436,"latitude":456,"longitude":457},{"@type":425,"streetAddress":459,"addressLocality":460,"addressRegion":461,"postalCode":462,"addressCountry":463},[1243,1244,1245],{"@type":466,"name":467},{"@type":466,"name":469},{"@type":466,"name":463},{"@type":472,"@id":473,"name":4,"image":1247,"description":6,"sameAs":1248,"url":200},{"@type":377,"inLanguage":366,"@id":475,"url":8,"contentUrl":8,"caption":4},[196],{"self":1250,"collection":1255,"about":1257,"author":1259,"replies":1261,"version-history":1264,"predecessor-version":1268,"wp:featuredmedia":1272,"wp:attachment":1275,"wp:term":1278,"curies":1285},[1251],{"href":1252,"targetHints":1253},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F29789",{"allow":1254},[483,484,485,486,487],[1256],{"href":490},[1258],{"href":493},[1260],{"embeddable":108,"href":496},[1262],{"embeddable":108,"href":1263},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcomments?post=29789",[1265],{"count":1266,"href":1267},19,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F29789\u002Frevisions",[1269],{"id":1270,"href":1271},39752,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F29789\u002Frevisions\u002F39752",[1273],{"embeddable":108,"href":1274},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia\u002F29830",[1276],{"href":1277},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia?parent=29789",[1279,1281,1283],{"taxonomy":516,"embeddable":108,"href":1280},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcategories?post=29789",{"taxonomy":519,"embeddable":108,"href":1282},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcollections?post=29789",{"taxonomy":180,"embeddable":108,"href":1284},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcoauthors?post=29789",[1286],{"name":525,"href":526,"templated":108},{"id":1288,"date":1289,"date_gmt":1289,"guid":1290,"modified":172,"modified_gmt":172,"slug":1292,"status":19,"type":113,"link":171,"title":1293,"content":1294,"excerpt":1296,"author":213,"featured_media":1297,"comment_status":20,"ping_status":20,"sticky":211,"template":5,"format":215,"meta":1298,"categories":1301,"collections":1302,"coauthors":1303,"class_list":1304,"acf":1306,"head_links":1307,"breadcrumbs":1308,"url_path":171,"search_description":1300,"featured_image_urls":1313,"authors":1345,"type_nicename":290,"reading_minutes":173,"category":1357,"yoast_title":1299,"yoast_meta":1360,"yoast_json_ld":1387,"_links":1438},29226,"2024-06-24T15:42:26",{"rendered":1291},"\u002F?p=29226","building-trustworthy-media-insights-through-model-validation",{"rendered":169},{"rendered":1295,"protected":211},"\u003Cdiv class=\"wp-block-image\">\n\u003Cfigure class=\"aligncenter size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1126\" height=\"470\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fmodel-validation.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-29236\"\u002F>\u003C\u002Ffigure>\u003C\u002Fdiv>\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"what-is-model-validation-and-why-is-it-important\">What is model validation and why is it important?\u003C\u002Fh2>\n\n\n\n\u003Cp>Model validation is the process of evaluating a statistical model’s performance and confirming the model achieves its intended purpose. It involves a series of techniques and assessments to determine how well a model predicts outcomes using new data that were not part of the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fbuiltin.com\u002Fdata-science\u002Ftrain-test-split\">training set\u003C\u002Fa>. This step is crucial and should not be overlooked because it provides confidence that the model will perform well on unseen data, thereby avoiding \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fwww.investopedia.com\u002Fterms\u002Fo\u002Foverfitting.asp\">overfitting\u003C\u002Fa>, where the model works well on training data but fails to generalise (or predict) to new data.\u003C\u002Fp>\n\n\n\n\u003Cp>In the context of \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Fmeasurement\u002F\">media effectiveness measurement\u003C\u002Fa>, model validation is particularly important as it ensures that the insights derived from the model about the impact of various media channels are accurate and actionable. Without thorough validation, any decisions based on a statistical model could lead to ineffective or even detrimental marketing strategies, resulting in wasted resources and missed opportunities for marketing optimisation. Proper validation helps in identifying and correcting biases, improving model robustness, and ultimately driving more effective media planning and resource allocation.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"model-validation-techniques\">Model validation techniques\u003C\u002Fh2>\n\n\n\n\u003Cp>When validating statistical models, several techniques can be employed to ensure the model performs well. One common technique is \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing\u002F#hold-out-testing-for-the-synthetic-control\">hold-out testing\u003C\u002Fa>, which involves splitting the dataset into two distinct parts: a training set and a testing set. A common split would be 80:20, or 70:30. The model is then trained on the training set and evaluated on the testing set. This approach is straightforward and easy to implement, providing a clear and unbiased estimate of the model’s performance on new data. The main limitation of hold-out testing is that it may not utilise the data as efficiently as cross-validation, especially with small datasets, and the results can be highly dependent on how the data is split.\u003C\u002Fp>\n\n\n\u003Cdiv class=\"wp-block-image\">\n\u003Cfigure class=\"aligncenter size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1420\" height=\"706\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fcross-validation.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-29234\"\u002F>\u003C\u002Ffigure>\u003C\u002Fdiv>\n\n\n\u003Cp>Another widely used technique is \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Flink.springer.com\u002Freferenceworkentry\u002F10.1007\u002F978-0-387-39940-9_565#:~:text=Definition,used%20to%20validate%20the%20model.\">cross-validation\u003C\u002Fa>. Cross-validation involves dividing the dataset into multiple subsets, or “folds”. The model is then trained on some folds while being tested on the other fold, a process that is repeated several times, with different folds used as the test set each time. This process is a similar concept to hold-out testing, but with the train\u002Ftest split occurring at different places each iteration. This technique is particularly beneficial as it maximises the use of available data and provides a robust estimate of the model’s performance. However, it can be computationally expensive, especially with large datasets.\u003C\u002Fp>\n\n\n\n\u003Cp>Each of these techniques has its benefits and limitations. Cross-validation offers a thorough validation process and maximises data use, but it can be computationally expensive. Hold-out testing is simple and unbiased but might not be efficient with smaller datasets. Choosing the appropriate validation technique depends on the specific context of the model and the available data, balancing between computational feasibility and the need for robust performance estimates.\u003C\u002Fp>\n\n\n\u003Cdiv class=\"wp-block-image\">\n\u003Cfigure class=\"aligncenter size-full\">\u003Ca rel=\"noreferrer noopener\" href=\"\u002Fmedia-solutions\u002Fmeasurement\u002F\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"3840\" height=\"540\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002FMeasurement_Banner_01.jpg?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-27690\"\u002F>\u003C\u002Fa>\u003C\u002Ffigure>\u003C\u002Fdiv>\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"model-validation-in-marketing-mix-modelling\">Model validation in Marketing Mix Modelling\u003C\u002Fh2>\n\n\n\n\u003Cp>Model validation is vital in \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fmedia-mix-modelling\u002F\">Marketing Mix Modelling (MMM)\u003C\u002Fa> because it confirms that the model accurately represents the relationship between marketing activities and sales, ensuring that the insights derived are actionable and reliable. Proper validation helps in identifying any inaccuracies, providing a clear understanding of how different marketing tactics influence outcomes. For MMM specifically, cross-validation and hold-out testing are the most frequently used techniques due to their ability to test model robustness and prevent overfitting.\u003C\u002Fp>\n\n\n\u003Cdiv class=\"wp-block-image\">\n\u003Cfigure class=\"aligncenter size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"1611\" height=\"611\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fmarketing-mix-model-and-sales-data.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-29235\"\u002F>\u003C\u002Ffigure>\u003C\u002Fdiv>\n\n\n\u003Cp>\u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fstatisticsbyjim.com\u002Fbasics\u002Fgoodness-of-fit\u002F\">Goodness of fit\u003C\u002Fa> is closely related to validation in statistical modelling. Goodness of fit measures how well a model’s predicted values match the observed data (visual above), serving as an initial check on model performance. Once the relevant goodness of fit checks are done, we introduce proper validation, testing the model’s performance on unseen data to ensure it generalises well beyond the sample used for estimation. Goodness of fit is therefore a key component of the broader validation process, ensuring the model is both accurate and generalizable, which is crucial for making reliable predictions and informed decisions in MMM.\u003C\u002Fp>\n\n\n\n\u003Cp>Overall, be sure to ask (along with other \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fquestions-to-ask-marketing-mix-modelling-vendors\u002F\">essential questions\u003C\u002Fa>) that your MMM vendor conducts these validations and goodness of fit techniques, as these methods provide you with confidence in your media investment decisions, ensuring that your marketing strategies developed are based on sound and validated models.\u003C\u002Fp>\n\n\n\n\u003Ch2 class=\"wp-block-heading\" id=\"model-validation-in-incrementality-testing\">Model validation in Incrementality Testing\u003C\u002Fh2>\n\n\n\u003Cdiv class=\"wp-block-image\">\n\u003Cfigure class=\"aligncenter size-full\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"952\" height=\"274\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fcausal-impact-incrementality.png?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-29233\"\u002F>\u003C\u002Ffigure>\u003C\u002Fdiv>\n\n\n\u003Cp>Model validation is crucial in \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing-media-planning\u002F#the-challenges-of-measuring-media-effectiveness\">incrementality testing\u003C\u002Fa> as, especially in the case of \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fwhat-is-synthetic-data\u002F\">synthetic control\u003C\u002Fa>-based testing, the model predictions essentially determine the outcome. As mentioned in \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing\u002F#deriving-the-control-groups\">previous blogs\u003C\u002Fa>, a synthetic control-based incrementality test boils down to comparing a “baseline” forecast (generated by a statistical model), in which the treatment is assumed absent, with the actual outcome. If the model doesn’t perform well on unseen data, the forecasts won’t accurately capture what would’ve happened in the absence of the treatment. \u003C\u002Fp>\n\n\n\n\u003Cp>With regards to validation techniques, cross-validation and hold-out testing are powerful and frequently implemented methods of assessing the predictive accuracy of your synthetic control. To help measure forecasting accuracy, the calculation of the \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fhelp.sap.com\u002Fdocs\u002FSAP_PREDICTIVE_ANALYTICS\u002F41d1a6d4e7574e32b815f1cc87c00f42\u002F5e5198fd4afe4ae5b48fefe0d3161810.html\">root-mean-square-error\u003C\u002Fa> (RMSE) and \u003Ca href=\"https:\u002F\u002Fwww.khanacademy.org\u002Fmath\u002Fstatistics-probability\u002Fdescribing-relationships-quantitative-data\u002Fregression-library\u002Fa\u002Fintroduction-to-residuals\">residual analysis\u003C\u002Fa> is recommended in addition to validation techniques. The RMSE is a measure of how far out, on average, our model predictions are when compared with historical train data. A small RMSE indicates good model performance. Additionally, residuals are defined as the difference between the predicted observations and the actual observations in the train data. Good model performance is implied when your residuals appear to be a sequence of random numbers, close to 0, that don’t follow any predictable pattern or trend over time. In other words, when your residuals closely mimic \u003Ca rel=\"noreferrer noopener\" href=\"https:\u002F\u002Fmachinelearningmastery.com\u002Fwhite-noise-time-series-python\u002F#:~:text=What%20is%20a%20White%20Noise,other%20values%20in%20the%20series.\">white noise\u003C\u002Fa>.\u003C\u002Fp>\n\n\n\n\u003Cp>Effective validation helps confirm that the \u003Ca rel=\"noreferrer noopener\" href=\"\u002Fblog\u002Fincrementality-testing\u002F#how-to-calculate-incrementality-and-lift\">observed lift\u003C\u002Fa> or impact is truly due to the marketing intervention and not due to forecasting inaccuracies or other external factors. This validation is essential to provide confidence in the incrementality estimates, thereby enabling you to make informed decisions about future campaigns, channels and budget allocations.\u003C\u002Fp>\n\n\n\n\u003Cfigure class=\"wp-block-image size-full\">\u003Ca rel=\"noreferrer noopener\" href=\"\u002Fget-in-touch\u002F\">\u003Cimg loading=\"lazy\" decoding=\"async\" width=\"3840\" height=\"540\" src=\"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F03\u002FGeneric_Banner_01.jpg?auto=compress%2Cformat&q=90\" alt=\"\" class=\"wp-image-27527\"\u002F>\u003C\u002Fa>\u003C\u002Ffigure>\n",{"rendered":170,"protected":211},29238,{"_yoast_wpseo_focuskw":5,"_yoast_wpseo_title":1299,"_yoast_wpseo_metadesc":1300,"inline_featured_image":211,"footnotes":5},"Building trustworthy media insights through model validation | Impression","Model validation is the process of evaluating a statistical model's performance and confirming the model achieves its intended purpose.",[219,220],[],[223],[1305,113,226,227,228,229,230,231,232],"post-29226",[],[],[1309,1310,1311,1312],{"path":185,"name":186},{"path":238,"name":239},{"path":135,"name":136},{"path":194,"name":169},{"thumbnail":1314,"medium":1315,"medium_large":1317,"large":1319,"1536x1536":1321,"2048x2048":1323,"post-thumbnail":1325,"12_col":1327,"default_landscape":1329,"default_portrait":1331,"6_col_and_margin_short":1333,"11_col_and_margin":1335,"9_col_rectangle":1337,"11_col":1339,"staff_image_696_600":1341,"rectangular_thumbnail":1343},{"url":174,"width":244,"height":244},{"url":1316,"width":70,"height":70},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=113&q=90&w=200",{"url":1318,"width":73,"height":23},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=432&q=90&w=768",{"url":1320,"width":251,"height":252},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1000&q=90&w=1892",{"url":1322,"width":79,"height":79},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=864&q=90&w=1536",{"url":1324,"width":82,"height":82},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=scale&h=1152&q=90&w=2048",{"url":1326,"width":85,"height":86},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1180&q=90&w=1860",{"url":1328,"width":88,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2880",{"url":1330,"width":91,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=800",{"url":1332,"width":65,"height":91},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=800&q=90&w=600",{"url":1334,"width":85,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=1860",{"url":1336,"width":95,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=3080",{"url":1338,"width":97,"height":98},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=1400&q=90&w=2150",{"url":1340,"width":100,"height":89},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=870&q=90&w=2640",{"url":1342,"width":64,"height":65},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=600&q=90&w=696",{"url":1344,"width":103,"height":104},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&fit=crop&h=300&q=90&w=750",[1346],{"id":213,"path":279,"display_name":4,"job_title":11,"description":6,"avatar_url":122,"acf":1347},{"full_bio":10,"job_title":11,"specialist_service_links":1348,"photo":1355},[1349,1351,1353],{"page":1350,"link_text":29},{"ID":15,"post_author":16,"post_date":17,"post_date_gmt":17,"post_content":5,"post_title":18,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":21,"to_ping":5,"pinged":5,"post_modified":22,"post_modified_gmt":22,"post_content_filtered":5,"post_parent":23,"guid":24,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":28},{"page":1352,"link_text":34},{"ID":32,"post_author":16,"post_date":33,"post_date_gmt":33,"post_content":5,"post_title":34,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":35,"to_ping":5,"pinged":5,"post_modified":36,"post_modified_gmt":36,"post_content_filtered":5,"post_parent":37,"guid":38,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":39},{"page":1354,"link_text":44},{"ID":42,"post_author":16,"post_date":43,"post_date_gmt":43,"post_content":5,"post_title":44,"post_excerpt":5,"post_status":19,"comment_status":20,"ping_status":20,"post_password":5,"post_name":45,"to_ping":5,"pinged":5,"post_modified":46,"post_modified_gmt":46,"post_content_filtered":5,"post_parent":37,"guid":47,"menu_order":23,"post_type":25,"post_mime_type":5,"comment_count":26,"filter":27,"url_path":48},{"ID":50,"id":50,"title":51,"filename":52,"filesize":53,"url":54,"link":55,"alt":5,"author":56,"description":5,"caption":5,"name":57,"status":58,"uploaded_to":23,"date":59,"modified":59,"menu_order":23,"mime_type":60,"type":61,"subtype":62,"icon":63,"width":64,"height":65,"sizes":1356},{"thumbnail":67,"thumbnail-width":68,"thumbnail-height":68,"medium":69,"medium-width":70,"medium-height":71,"medium_large":72,"medium_large-width":73,"medium_large-height":74,"large":75,"large-width":76,"large-height":77,"1536x1536":78,"1536x1536-width":79,"1536x1536-height":80,"2048x2048":81,"2048x2048-width":82,"2048x2048-height":83,"post-thumbnail":84,"post-thumbnail-width":85,"post-thumbnail-height":86,"12_col":87,"12_col-width":88,"12_col-height":89,"default_landscape":90,"default_landscape-width":91,"default_landscape-height":65,"default_portrait":92,"default_portrait-width":65,"default_portrait-height":91,"6_col_and_margin_short":93,"6_col_and_margin_short-width":85,"6_col_and_margin_short-height":89,"11_col_and_margin":94,"11_col_and_margin-width":95,"11_col_and_margin-height":89,"9_col_rectangle":96,"9_col_rectangle-width":97,"9_col_rectangle-height":98,"11_col":99,"11_col-width":100,"11_col-height":89,"staff_image_696_600":101,"staff_image_696_600-width":64,"staff_image_696_600-height":65,"rectangular_thumbnail":102,"rectangular_thumbnail-width":103,"rectangular_thumbnail-height":104},[1358,1359],{"id":219,"name":125,"path":124},{"id":220,"name":136,"path":135},[1361,1362,1363,1364,1366,1367,1368,1369,1370,1371,1372,1373,1375,1377,1379,1380,1381,1382,1383,1384,1385,1386],{"property":296,"content":1299},{"property":298,"content":1300},{"name":300,"content":301},{"property":199,"content":1365},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fbuilding-trustworthy-media-insights-through-model-validation\u002F",{"name":305,"content":306},{"property":308,"content":309},{"property":311,"content":1299},{"property":313,"content":1300},{"property":315,"content":1365},{"property":317,"content":318},{"property":320,"content":321},{"property":323,"content":1374},"2024-06-24T15:42:26+00:00",{"property":326,"content":1376},"2024-06-24T15:44:16+00:00",{"property":329,"content":1378},"https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fopen-graph-model-validation.png?auto=compress%2Cformat&q=90",{"property":332,"content":333},{"property":335,"content":336},{"property":338,"content":60},{"property":180,"content":4},{"property":341,"content":342},{"property":344,"content":1378},{"property":346,"content":347},{"property":349,"content":347},{"@context":351,"@graph":1388},[1389,1400,1406,1407,1413,1417,1435],{"@type":290,"@id":1390,"isPartOf":1391,"author":1392,"headline":169,"datePublished":1374,"dateModified":1376,"mainEntityOfPage":1393,"wordCount":1394,"publisher":1395,"image":1396,"thumbnailUrl":1398,"articleSection":1399,"inLanguage":366},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fbuilding-trustworthy-media-insights-through-model-validation\u002F#article",{"@id":1365},{"name":4,"@id":357},{"@id":1365},975,{"@id":361},{"@id":1397},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fbuilding-trustworthy-media-insights-through-model-validation\u002F#primaryimage","https:\u002F\u002Fimages.impression.co.uk\u002F2024\u002F06\u002Fgoogle-deepmind-jVZ_BKzDOJg-unsplash.jpg?auto=compress%2Cformat&q=90",[125,136],{"@type":368,"@id":1365,"url":1365,"name":1299,"isPartOf":1401,"primaryImageOfPage":1402,"image":1403,"thumbnailUrl":1398,"datePublished":1374,"dateModified":1376,"description":1300,"breadcrumb":1404,"inLanguage":375},{"@id":370},{"@id":1397},{"@id":1397},{"@id":1405},"https:\u002F\u002Fwww.impressiondigital.com\u002Fblog\u002Fbuilding-trustworthy-media-insights-through-model-validation\u002F#breadcrumb",{"@type":377,"inLanguage":366,"@id":1397,"url":1398,"contentUrl":1398,"width":378,"height":379},{"@type":381,"@id":1405,"itemListElement":1408},[1409,1410,1411,1412],{"@type":384,"position":385,"name":186,"item":386},{"@type":384,"position":388,"name":239,"item":389},{"@type":384,"position":391,"name":136,"item":651},{"@type":384,"position":394,"name":169},{"@type":396,"@id":370,"url":386,"name":318,"description":5,"publisher":1414,"inLanguage":366,"potentialAction":1415},{"@id":398},{"@type":400,"target":1416,"query-input":404},{"@type":402,"urlTemplate":403},{"@type":406,"@id":398,"name":318,"url":386,"logo":1418,"image":412,"sameAs":1419,"@context":417,"email":418,"foundingDate":419,"legalName":420,"vatID":421,"description":422,"telephone":423,"address":1420,"location":1421,"areaServed":1431},{"@type":377,"inLanguage":366,"@id":408,"url":409,"contentUrl":409,"width":410,"height":411,"caption":318},[414,415,416],{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},[1422,1425,1428],{"@type":433,"name":434,"geo":1423,"address":1424,"photo":412,"telephone":440},{"@type":436,"latitude":437,"longitude":438},{"@type":425,"streetAddress":426,"addressLocality":427,"addressRegion":428,"postalCode":429,"addressCountry":430},{"@type":433,"name":442,"geo":1426,"address":1427,"photo":451,"telephone":452},{"@type":436,"latitude":444,"longitude":445},{"@type":425,"streetAddress":447,"addressLocality":448,"addressRegion":449,"postalCode":450,"addressCountry":430},{"@type":433,"name":454,"geo":1429,"address":1430},{"@type":436,"latitude":456,"longitude":457},{"@type":425,"streetAddress":459,"addressLocality":460,"addressRegion":461,"postalCode":462,"addressCountry":463},[1432,1433,1434],{"@type":466,"name":467},{"@type":466,"name":469},{"@type":466,"name":463},{"@type":472,"@id":473,"name":4,"image":1436,"description":6,"sameAs":1437,"url":200},{"@type":377,"inLanguage":366,"@id":475,"url":8,"contentUrl":8,"caption":4},[196],{"self":1439,"collection":1444,"about":1446,"author":1448,"replies":1450,"version-history":1453,"predecessor-version":1456,"wp:featuredmedia":1460,"wp:attachment":1463,"wp:term":1466,"curies":1473},[1440],{"href":1441,"targetHints":1442},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F29226",{"allow":1443},[483,484,485,486,487],[1445],{"href":490},[1447],{"href":493},[1449],{"embeddable":108,"href":496},[1451],{"embeddable":108,"href":1452},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcomments?post=29226",[1454],{"count":154,"href":1455},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F29226\u002Frevisions",[1457],{"id":1458,"href":1459},29240,"\u002Fwp-json\u002Fwp\u002Fv2\u002Fposts\u002F29226\u002Frevisions\u002F29240",[1461],{"embeddable":108,"href":1462},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia\u002F29238",[1464],{"href":1465},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fmedia?parent=29226",[1467,1469,1471],{"taxonomy":516,"embeddable":108,"href":1468},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcategories?post=29226",{"taxonomy":519,"embeddable":108,"href":1470},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcollections?post=29226",{"taxonomy":180,"embeddable":108,"href":1472},"\u002Fwp-json\u002Fwp\u002Fv2\u002Fcoauthors?post=29226",[1474],{"name":525,"href":526,"templated":108},{"title":1476,"script":1477,"meta":1491,"link":1504},"Harry Brace | Director, Head of Solutions | Impression",[1478],{"type":1479,"key":1480,"innerHTML":1481},"application\u002Fld+json","yoast-json-ld",{"@context":351,"@graph":1482},[1483,1488],{"@id":473,"@type":472,"name":4,"url":-1,"sameAs":1484,"image":1485,"description":6,"jobTitle":11,"knowsAbout":1486,"hasCredential":5,"gender":5,"worksFor":1487},[196],{"@type":377,"inLanguage":366,"@id":475,"url":8,"contentUrl":8,"caption":4},"Web Analytics, Tag Management, Google Marketing Platform",{"@id":398},{"@type":406,"@id":398,"name":318,"url":386,"logo":1489,"image":412,"sameAs":1490,"@context":417,"email":418,"foundingDate":419,"legalName":420,"vatID":421,"description":422,"telephone":423},{"@type":377,"inLanguage":366,"@id":408,"url":409,"contentUrl":409,"width":410,"height":411,"caption":318},[414,415,416],[1492,1493,1494,1496,1497,1499,1500,1502,1503],{"name":300,"content":301},{"property":305,"content":366},{"property":308,"content":1495},"profile",{"property":311,"content":1476},{"property":315,"content":1498},"https:\u002F\u002Fadmin.impressiondigital.com\u002Fabout\u002Four-team\u002Fharry-brace\u002F",{"property":317,"content":318},{"property":329,"content":1501},"https:\u002F\u002Fimages.impression.co.uk\u002F2023\u002F10\u002FHarry-Brace2.png?auto=compress&fit=fit&fm=jpg&h=500&q=90&w=500",{"name":341,"content":342},{"name":349,"content":347},[]]