Marketing Mix Modeling (MMM) is an important method for optimizing marketing mix and attribution in a world where cookies will soon no longer exist. Marketing Mix Modeling has already been successfully applied by companies that want to continue to play a role in this cookie-free world. Due to this increase in MMM applications, we also see in practice that companies often fall into the same pitfalls. Pitfalls that are not (yet) prevented by intelligence within tooling. With this article I hope to save you from these pitfalls.
The 7 most important pitfalls at MMM
1. Not validating the MMM project
Pitfall #1. Many Marketing Mix Modeling projects are set up on a grand scale and with high expectations. An external agency is called in and an internal project team is set up. So many direct and indirect costs are incurred in advance without validating the MMM project in itself (“Think big, start bigger”). Validating this in advance is important to investigate whether your organization is able to derive significant benefits from the MMM project and to avoid the further pitfalls mentioned in this article as much as possible.
So, “think big, start small”. First set up a small-scale MMM project independently with the help of your own employees or graduates and simple (free) tooling. For the latter you can use, for example, the Key Influencer (multivariate linear regression) analysis within Power BI, or with R or Python. Validate for usability and applicability of the results and their possible acceptance and integration within your organization. Additional advantage of this approach; you wouldn’t be the first to extend this validation to a serious in-house MMM project.
2. Using insufficient or incorrect data
Pitfall #2. Marketing Mix Modeling requires a lot of historical data at a low granular level. An example of such data is the number of newsletter subscriptions per campaign, per ad group, per advertisement, per landing page and per day in the past 3 years. This doesn’t seem like much, but if you had 4 continuous campaigns running for your newsletter subscriptions in these 3 years, each consisting of 5 ad groups with 10 ads per ad group that lead traffic to 2 different landing pages, then this generates a table with 4 x 5 x 10 x 2 x 3 x 365 = 438,000 unique lines. And that is only for your newsletter lead campaigns. Not yet for your other marketing goals, campaign data, organic media, customer data, customer contact data, transaction and/or checkout data, product data, pricing data, stock and distribution data, etc.
Much historical data is aggregated to a high-level level such as totals or averages. Sometimes this happens automatically and unnoticed (for example within standard Google Analytics reports), often this choice was made in the past to control data storage costs. In any case, aggregated data is of little use within Marketing Mix Modeling. The unavailability of the correct historical data with the correct granularity can therefore already nip your MMM project in the bud.
Of course it can also happen that the required data is not available and available at all. In that case, you should ask yourself whether an MMM project is even suitable for your organization at all.
3. Ignoring External Influences
Pitfall #3. External influences also influence the behavior of your customers and therefore your results. For Marketing Mix Modeling, therefore, external data is required in addition to our own historical data. Think of relevant data from competitors (their product, pricing and promotion data), weather and traffic, holidays and public holidays, macroeconomic data etc. Much of this data is available almost free of charge.
4. Not applying anomaly detection
Pitfall #4. Large unrepresentative deviations in the data, also called anomalies or outliers, can seriously disrupt the outcomes of your Marketing Mix Modeling project. It is therefore advisable to detect and repair these anomalies already during the validation of your MMM project and the validation of your data. You can easily do anomaly detection in Power BI, R or Python, among others.
You can repair anomalies by replacing deviating peaks and valleys with, for example, averages or null values (data infusion), which you also do when data is missing. The causes of anomalies and missing values are often technical in nature, so it is also advisable to repair these causes to prevent recurrence.
5. Ignoring the killer deal effect
Pitfall #5. That one big promotion, which you normally don’t do very often, also has a strong influence on your Marketing Mix Modeling model. So consider in advance whether or not you want those promotion(s), with the associated data, to be part of your MMM project, or whether you want to treat them as one of the anomalies mentioned above.
6. Selecting the wrong model
Pitfall #6. The foundation of Marketing Mix Modeling is, you guessed it, the Marketing Mix Model. In addition to the aforementioned multivariate linear regression model, this can also be a SCAN*PRO model, for example.
7. Using the wrong input variables
Pitfall #7. The two most important input variables within your Marketing Mix Modeling project are your base line and ad stock. If you use the wrong values here, you will never get the right results, or you will misinterpret those results.
The goal of your Marketing Mix Modeling project is to find out which combinations of variables and their values increase your results. It sounds very logical, but for that it is necessary to define your ‘base line’; the results you achieve (or would achieve) without any marketing efforts. The easiest way to do that is by taking your organic and direct results (i.e. without campaigns and promotions) as a starting point.
Ad stock is the long-term or delayed effect of advertising on consumer buying behaviour. Precisely because within Marketing Mix Modeling projects you start from time series data at a low granular level, it is important to monitor the effect of your branding and generic campaigns and promotions on the behavior of your customers over the longer term. The ad stock calculation is already included in the articles referred to earlier in this article.
Conclusion
Marketing Mix Modeling (MMM) is an important method for optimizing marketing mix and attribution in a world where cookies will soon no longer exist. The road to MMM success contains pitfalls, but they are surmountable. If you have any questions about this, ask them in the comment box at the bottom of this article or contact me.