Marketing Mix Modeling Software - Build vs. Buy

Marketing Mix Modeling (MMM) has become a widely adopted practice among consumer goods companies over the past two decades, with many Fortune 500 companies making it an essential component of their marketing planning. As technology has evolved and new online channels have emerged, Multi-Touch Attribution has gained popularity. However, with the decline of cookie-based tracking methods for consumer behavior, the focus has shifted to MMM. This has led to a resurgence of Marketing Mix Modeling software, also supported by the development of modern data pipelines, advanced statistical approaches like machine learning or Bayesian methods, and open-source options such as Robyn from Facebook (Meta) and LightweightMMM from Google, which has made MMM more affordable and easier to execute.

Brands are realizing the need for new tools and for expertise to accurately assess their marketing effectiveness. This presents a common dilemma: whether to build a customized Marketing Mix Modeling (MMM) solution in-house or seek out third-party assistance. To facilitate this decision-making process for you, we've compiled a guide outlining the benefits and drawbacks of each approach, including the involvement of specialized companies or the application of ready-to-use software solutions.

Questions addressed in this article

• What resources are needed to build a Marketing Mix Model in-house?
• How long will it take to build MMM in-house?
• What is the cost of building Marketing Mix Modeling in-house?
• How much will it cost to buy Marketing Mix Modeling software?
• What to look out for when buying Marketing Mix Modeling Software?

How to Build a Marketing Mix Modeling In-House?

Rule of thumb: You can't build Marketing Mix Modeling in-house without a team of good data scientists, data analysts, and data engineers with the following skills:

• Good knowledge of the market (or at least experience in analyzing various markets or branches) in order to identify possible influencing factors
• Basic knowledge of how media channels work
• Good overview of possible statistical methods as well as their advantages and disadvantages, and experience with their application
• Ability to use analytical software or scripting languages
• Capability to connect available external data sources and merge the data with the internal media data
• Experience in data preparation (merging, transforming, and assessing data quality) and ability to recognize correlations

Building an internal data intelligence team is often not easy, because professionals in this field are in high demand and well paid, as listed further down in the article. This should be taken into account.

Building your own Marketing Mix Modeling solution can be a challenging task due to the complexity of your company's marketing activity. We would like to highlight a few specific difficulties:

• The first challenge is to specify the model correctly. That means, all the relevant influencing factors have to be included, but overfitting has to be avoided. Moreover, sufficient data points for individual influencing factors must be available.
• The presence of multicollinearity, or correlation between marketing channels, can make it difficult to create accurate models and has to be taken into account. This happens when marketers simultaneously book different media channels (which of course often happens) or when several external events (e.g. holidays, Black Friday, etc.) occur at the same time.
• Non-linear correlations, in particular time-delayed media effects like the adstock effect and saturation, have to be mapped, and the same applies to interaction effects.
• It is important to take into account long-term brand development as well as the relationship between media budget and media performance, which makes budget allocation in particular more complex.
• Budget allocation is not an easy task, as it requires knowledge of optimization algorithms and involves definition of allocation procedures.
• Marketing Mix Modeling becomes more complex if it covers multiple regions and/or multiple brands, which requires a hierarchical approach.
• As data quality and availability of all the relevant data in general is the key to successful Marketing Mix Modeling, it's crucial to ensure that data pipelines are automated.

Furthermore, the model needs to comply with a long checklist of necessary features, charts, tables, and recommendations for decision-making. At MMT, we believe that to provide comprehensive insights into your marketing efforts, a successful Marketing Mix Model should include the following features:

• Attribution modeling: Quantifiying the contribution of media      channels to the target metric and their efficiency (ROI analysis).
• Recommendation: Optimizing the media mix
• Predictive analytics: Creating forecasts for campaign scenarios
• Hierarchical approach: Generating individualized models for different regions or brands, if required.

Marketing Mix Modeling is a complex process that requires time and resources. According to Facebook (Meta), it takes 12-22 weeks to manually build the initial model, not taking into account continuous development and maintenance work required for automation. Smaller companies can achieve a "good enough" model in just a few weeks, but for larger organizations, the typical timeline ranges from several months to more than half a year.

Assuming a team of two data scientists working full-time, it could take approximately six months to build and automate an initial model. The average salary of a data scientist in Germany in 2023 is around €65,000 according to job platforms, with potentially higher salaries in major cities like Berlin, Hamburg or Munich. Once the initial model is finalized, ongoing maintenance will require at least one day every two weeks as well as a four-week update cycle each year.
The costs of data procurement, connection, and automation must also be taken into account, and you might even need to hire a dedicated data integration engineer for this purpose.

What to look for when buying Marketing Mix Modeling Software?

Marketing Mix Modeling (MMM) has become a popular tool in the marketing field, with various vendors offering their own approaches. It’s important to understand the scientific basis behind individual approaches in order to select the one that best suits your needs. One major differentiation within MMM is between frequentist and Bayesian methods. Traditional MMM relied on ordinary least squares (OLS) regression, whereas Facebook's Robyn algorithm utilizes a more modern ridge regression approach. Alternatively, Google has adopted Bayesian methods for their modeling techniques, which allows for greater flexibility in incorporating previous knowledge into the model and producing more realistic outcomes.

The biggest difference in cost and maintenance will depend on your decision to either use the standard automated software or customize your MMM solution. Most companies offering Marketing Mix Modeling can build something custom for your business, if needed. A model tailored to the company's own needs can be capable of answering exactly the right questions instead of providing general guidelines. In addition, you can receive advice regarding the implementation of the entire analysis cycle and the interpretation of the results. However, the manual approach requires a lot of trust because MMM is susceptible to human bias. Automated tools can use modern data pipelines to update the model regularly, so decisions can be made more often without having to repeat the analysis. Automated models are also able to automatically generate budget allocation recommendations, which makes them far more actionable.

For us at MMT, it is crucial that advertisers could turn their MMM insights into action. Therefore our Marketing Mix Modeling platform provides different media mix scenarios that could be directly adapted to media plans within the same platform. Our solution is a self-service tool, giving advertisers the possibility to model on their own to really understand how the MMM works and thus not rely on trust. This way you have full control over the modeling process and build up the expertise in your own company.

When it comes to selecting a vendor, the following questions might help you find the right approach for your company:

• Do they offer an off-the-shelf solution or will they build a custom MMM tool?
• Have they already worked with similar companies and thus gained industry knowledge?
• How much media expertise do they have?
• What methods do they use? Linear regression, ridge regression or Bayesian methods?
• What’s the cost of the initial model building?
• What are the ongoing costs associated with maintenance / running the model regularly?
• Do they offer after-sales services?
• Do they have the key features mentioned above?
• Do they allow different levels of automation?
• Does the level of involvement required from your company match your own skills/available resources?
• Can they explain the applied methods and possible outputs in a comprehensible way?
• Will they give useful, competent, and transparent advice on how to answer your specific questions based on the modeling insights?

Once you've made that decision whether you want to use scalable software or have the provider build something custom for you, it's essential to consider in detail what the model will do. This can be challenging, given that there is no one-size-fits-all solution in this emerging field. Some things to keep in mind include the law of diminishing returns (which states that channels may become saturated at higher spends) and the adstock effect (the delayed impact of advertising on performance). Finally, it’s important to incorporate all relevant key features into your model to reflect diverse influence factors and deal with trends and seasonality. Choosing more advanced approaches that account for changes in channel performance over time or for the variance in correlations for different products or regions can significantly increase the accuracy of your model's outputs.

Conclusion

Quite often the answer to the question of whether a company should set up marketing mix modeling on its own or buy already existing software starts with "it depends". The key factor is the resources available. Building a model requires expertise and takes time. There is a lot of room for error in the calculations, which can lead to high losses if the marketing budget is used in a suboptimal way. And for marketers to really take advantage of the valuable insights, they need smart functionalities, as well as domain and media expertise. If it's your first experience with MMM or you have doubts about certain aspects, it might be worthwhile to involve a consultant/MMM specialist to guide you on your way and help you avoid unnecessary efforts and wrong marketing decisions.

Sources

• https://www.stepstone.de/gehalt/Data-Scientist.html
• https://en.wikipedia.org/wiki/Marketing_mix_modeling
• https://www.glassdoor.de/Geh%C3%A4lter/data-scientist-gehalt-SRCH_KO0,14.htm