Marketing Mix Modelling (MMM) is a powerful statistical tool which aims to understand the impact of various marketing activities on sales and other key performance metrics. At the heart of MMM lies the need for high-quality, accurate data containing multiple variables such as advertising spend, promotions, and external marketing factors. Therefore, the reliability and completeness of the data in MMM is crucial for generating actionable insights and making informed decisions in today’s competitive marketplace.
Regardless of your objectives, all MMM data should meet the following specifications:
- Of time series format.
- Contains approximately 3 years of observations.
- As granular as possible, daily is preferred.
- As accurate as possible, not subject to errors and/or missing observations.
- Data boundaries must align, if you’re conducting a US-only MMM, make sure your data is also US-only.
Sales data
In Marketing Mix Modeling, the decomposition of sales data serves as the focal point of analysis, representing the outcome that the model seeks to predict. The accuracy and reliability of your sales data is paramount, as any inaccuracies or inconsistencies can significantly impact the model’s predictive power. The following considerations are crucial when collecting your sales data:
- Online only sales or overall sales. (Usually overall sales is preferred)
- B2B, B2C or both. (This will depend on the specific analysis)
- Product groups. (This will depend on the objectives of analysis)
- Sales volume or sales value. (Usually value is best for generating ROI insights)
- Source (we advise raw first party financial data).
Marketing data
Marketing Mix Models utilise marketing data to assess the influence of various marketing activities on sales. Consequently, the accuracy of your marketing data significantly affects the model’s estimation capabilities. In MMM, it is crucial to incorporate both your digital and offline marketing data.
Digital marketing data
The most obvious data requirement for MMM is digital marketing data, but deciding which metric to use that best describes the performance of your marketing channel/campaign can be challenging. All MMM’s require marketing spend data as, even if spend data isn’t used for parameter estimation, it’s required to produce ROI estimates and to help guide budget forecasts.
Additionally, it is common for some MMM vendors to use engagement metrics such as impressions or clicks for parameter estimation, meaning that the contribution outputs produced by the model are derived using engagement metrics as opposed to spend.
| Metric | Benefit |
|---|---|
| Spend | Offers a direct measure of investment in marketing channels, allowing for straightforward comparisons of resource allocation, saturation and ROI across different channels. |
| Engagement metric (impressions, clicks…) | Offers a more nuanced understanding of consumer behaviour, by capturing the level of interaction and interest generated by marketing activities. |
Ultimately, the choice between engagement metrics and spend for parameter estimation depends on the specific objectives and priorities of the marketing analysis, with each approach offering unique insights into marketing performance.
Offline marketing activity
Incorporating offline marketing activity is essential in MMM because it provides a holistic view of how all your marketing efforts impact business outcomes. While digital marketing channels often receive significant attention due to their trackability and measurability, offline channels such as TV advertising, print media, outdoor campaigns and even affiliate activity still play a significant role in reaching and influencing consumers. Some examples and their associated metrics include:
| Variable | Metrics |
|---|---|
| TV | Gross rating points (GRP’s). |
| Circulation, readership, gross spend | |
| OOH (out of home) | Net spend, potential reach. |
Calibration data
Calibration in MMM refers to the process of fine-tuning the model parameters to ensure that the predicted outcomes align closely with actual observed data from incrementality tests, thereby enhancing the model’s accuracy and reliability. Provided you’ve recently performed an incrementality, incorporating the following elements of your incrementality test can help the model to provide greater causal insights.
- The time & date at which the test took place.
- The nature of the treatment (examples include switching advertising on/off, increasing/reducing spend by x%).
- The observed impact.
Control data
In Marketing mix modelling, control data refers to the variables or factors that are not directly influenced by marketing activities and are used as benchmarks or reference points for comparison. These variables typically include factors such as seasonality, economic conditions, significant events, and peak selling seasons that may impact sales. By incorporating control data into the model, we can isolate the impact of marketing efforts from other external influences. The following sections outline some common variables to include in your control data.
Pricing information
The aim of including a price-related control variable is to help account for changes in pricing dynamics and customer purchasing behaviour which may affect sales independently of marketing efforts. This ensures that changes in the sales attributed to marketing activities are not confounded by changes in price, providing a more accurate assessment of marketing effectiveness. The following are some commonly used pricing variables in MMM:
| Variable | Benefit |
|---|---|
| Average order value (AOV) | Allows greater understanding of how changes in marketing strategies impact customer spending behaviour. |
| Market price index (MPI) | Allows you to account for fluctuations in prices of goods or services, which may affect sales independently of marketing efforts. |
Seasonality
Seasonality refers to the recurring and predictable patterns or fluctuations in data that follow a regular and periodic cycle over time. Seasonality data is usually synthetically derived by the MMM practitioner using various methods.
Care must be taken when accounting for seasonality in MMM as marketing activity performs better when demand is high! In some cases, you find that controlling for seasonality will under-credit your marketing performance when your sales peak. If you find yourself in this position, you may wish to consider incorporating time variant parameters into your marketing mix model.
Events
Accounting for significant events such as promotions or seasonal holidays is essential in MMM because these events can impact consumer behaviour and purchasing patterns. By incorporating these events into the model, you can better understand the relationships between marketing activities and sales performance without the confounding influence of significant events. For this reason, it is essential that you keep track of any promotion data. Some examples of event data include the following:
| Event | Data required |
|---|---|
| Promotions | Nature of the promotion (x% off), the start and end date. |
| Public holidays | The date of the public holiday. |
| New product launches | The date of the new product launch. |
| Significant business changes | Nature of the change and the relevant dates at which this change occurred. |
| COVID-19 | Dates of lockdown periods. |
Trend & Baseline sales
Trend in data refers to the long-term direction or pattern of change in a variable over time. It represents the overall tendency of the data to increase, decrease, or remain relatively stable over an extended period. Similarly to seasonality, trend data is usually derived by the MMM practitioner using a number of different methods.
Baseline sales in MMM simply refers to how sales would behave in absence of all other variables. In MMM this output is usually referred to as the “intercept”. For most MMM’s, this intercept is modelled as constant, however at Impression we believe that baseline sales should be allowed to fluctuate over time if the data suggests so. To allow baseline sales to fluctuate overtime we treat the intercept as time-variant, in doing this we simultaneously aim to account for trends.
Macroeconomics
It is essential to include macroeconomic variables in MMM as they help account for external factors that can significantly influence consumer behaviour and marketing performance. Understanding the influence that purchasing power, disposable income and customer confidence has on demand enables you to isolate the impact of your marketing from broader economic trends, leading to more precise and actionable insights. The following are some examples of macroeconomic factors to include in MMM:
| Variable | Benefit |
|---|---|
| Customer price index (CPI) | Captures shifts in consumer spending patterns |
| Unemployment rate | Captures consumer confidence and disposable income. |
| Exchange rates | Captures the fluctuations in the cost of imported goods. |
Literally anything else
Each business is different and has unique complexities associated with it. When thinking about MMM, if you’re modelling sales, make sure you consider as many factors as possible that could influence sales. Even if some factors were found to have very little or no impact by the MMM, providing as much information as possible prior to modelling is essential!
Although we recommend providing as much data as possible, we do not recommend building an MMM and throwing all your available data at it. Including all of the above control data in less complex models might be overkill. Too many variables may not only lead to model overfitting, but also multicollinearity if many of these variables behave similarly. The art of a good MMM is therefore variable selection, which is where we as skilled practitioners step in to overcome this and many other important considerations for MMM.
Budgets
Provision of budgets is crucial for MMM because it provides essential context and constraints for evaluating the effectiveness of marketing activities. This is particularly useful when performing budget allocation in MMM, as knowledge of budgets will constrain the budget optimiser, enabling you to produce realistic forecasts to produce data-driven insights that you can actually implement.
