Some ideas on directions toward which Marketing Mix Modeling Software-as-a-Service Platforms should be developed:
- PROBLEM: Small cost variability (=spends remain stable over time and do not fluctuate enough to provide a range of values for the model detect significant associations when a campaign or channel’s spend goes up or down) in ‘set and forget’ type of campaigning causes problems for discerning the connections between campaigns/channels and revenue.
- SOLUTION: When this happens, a SaaS tool needs to recommend a *policy* for varying the budgets adequately to produce this variability. This is somewhat similar (conceptually) to the idea of sampling policy in active learning—-how to select samples for annotation? The policy is based on making strong changes to budget levels (e.g., +20%, -20%—-actual values to be determined in a principled manner) for a given test period.
- ANOTHER PROBLEM: Data imbalance; e.g., TikTok Ads has 10% of the budget, Google Ads 90%. It’s extremely difficult to make models work with highly imbalanced inputs, not only because of statistical reasons but also because platforms may require a certain scale before they start working. Again, the SaaS platform should boldly recommend policies for a given test period, not just take observational data and hope that marketers are testing correctly (which they certainly are not!).
- SOLUTION: Budget parity could serve as a strategy for tackling imbalance.
- SELLING MENTALITY, NOT A TOOL: The buyer needs to have tolerance for failure. Most tests fail! (=produce worse, not better performance). So, a digital marketing organization needs to be willing to accept failure. Otherwise, there’s no incentive to take risks, which is a necessary mentality to make big improvements. (If an MMM tool only produces minor improvements, there’s little point in using it—-just apply whatever strategies without such tool and you get approximately similar results, right?)
So, an MMM SaaS tool should not only observe but it should actively suggest and, if suggested policy is accepted by the human manager, implement it in the ad platforms!
I think this step from observations to interventions is necessary to unleash the potential of MMM.