How Marketing Mixed Modeling Can Improve Your ROI

The Limitation of Attribution Models

Revealing the true effect of any marketing activity on consumer behavior is a hard task. The different attribution models used today do not account for the complex journey a consumer takes before making a purchase, which often involves multiple touch points — offline & online alike. Unfortunately, these models — from Single-Touch Attribution to Multi-Touch Attribution — fail to detect the causal effect (positive or negative) between the different marketing activities and the consumer behavior. In addition, changes in privacy regulations significantly restricts the ability to track users across platforms, reducing the effectiveness of these models even more. These limitations highlight the need for advanced methods that can capture the actual effects of each marketing activity on your KPIs.

“One day, without really knowing how, a decision is made.” (Milan Kundera, Unbearable Lightness of Being)

When making marketing decisions you want to know what would have happened if you did things differently. For instance, the true effect of your Facebook ads can be determined by comparing your current KPIs to a hypothetical scenario where you’ve eliminated this channel while maintaining all others unchanged. This comparison reveals the actual impact of your spending on FB, which is very different from what click-through attribution provides.

Now imagine a metaphorical golden fish that could tell you how decreasing your Facebook budget by 10% wouldn’t really affect your KPIs, this can save a lot of money! Or it might suggest reallocating 5% of your budget from Google Ads to CTV ads can improve your KPIs. However, directly measuring this effect is impossible. Testing methods like A/B testing and geo-tests offer valuable insights, but they come with significant costs and fail to capture the entire picture. Ideally, we’d see how shifting budgets across all channels simultaneously affects your KPI, but in practice, we’re stuck with the choices we make without being able to know what would have happened if we did things differently.

Marketing Mix Modeling to the Rescue

The answer lies in Marketing Mix Modeling (MMM). This approach uses a sophisticated model that analyzes extensive historical data and uses it to estimate the true effect each channel has on your KPIs. To do this, the model breaks down the KPI, whether daily or weekly, into its contributing factors. This includes the organic baseline of sales, plus the influence of control variables like weather on trends. It then adds the impact of each marketing channel, factoring in expenditure, impressions, clicks, and even more nuanced aspects like lagged effects and saturation (where the impact of spend diminishes after a certain point). The model might also adjust for seasonal variations, recognizing that the effectiveness of ads can improve when demand is high (you don’t wanna advertise ski vacations in June). By aggregating and analyzing data over time, it deconstructs these factors to get the actual contribution of each channel to the KPI.

Building an MMM

In building an MMM, there are different tools in the market to choose from, each relying on different methods and assumptions. Google’s Meridian, for example, uses a Bayesian approach, which is usually great at quickly adapting to new data and incorporating domain knowledge. On the other side, Meta’s Robyn uses a frequentist method, focusing on stable, long-term data but might be slower to respond to quick changes in the market.

The key to making an MMM work well is strategically selecting variables that truly reflect your business’s dynamics, requiring a mix of causal inference and in-depth business knowledge. This includes accounting for unique factors like trending events or competitor actions. Also, running and incorporating test results can help calibrate the model and improve its accuracy even more.

After carefully incorporating variables and fine-tuning the parameters, you will have a model that captures the effect each marketing dollar has, and allows you to explore what other allocations can give you. Now you can shift your budget towards the most profitable areas. This strategic allocation hinges on a deep understanding of how different marketing efforts contribute to overall performance, guiding more informed and effective budgeting decisions.