How to use MMM for planning and forecasting

Marketers hate planning season. It can often feel like an impossible task: you’re being asked to predict the future in an industry where consumer sentiment (not to mention channel performance) can change suddenly and where multi-touch attribution tools (MTA) can’t be used for reliable predictions.

Over the last few years marketing mix modeling (MMM) has grown in popularity as a solution to help marketers measure marketing effectiveness in the face of signal-loss related to privacy regulations and policies like Apple’s App Tracking Transparency (ATT) initiative.

It turns out that MMMs are useful beyond just measuring marketing performance. When built and managed correctly, a good MMM isn’t just used to generate interesting insights but rather should be the most important tool in your toolbox for monthly, quarterly, and yearly marketing budget planning.

Using MMM for Forecasting and Planning

The output of an MMM (if it is well-designed) includes estimates for the causal relationship between the variables in the model and the business KPI (often something like “revenue”). These causal estimates can be used to make forward-looking forecasts about what will happen if you vary your marketing mix into the future.

That means that MMMs can (and should!) be used for forecasting and planning. It should be a relatively straightforward process to take a planned budget from a marketer that shows how much they’re planning to spend on each channel into the future and then apply the relevant estimates from the MMM in order to get a forecast.

In the best-case scenario, there’s an interactive tool that marketers can use to seamlessly upload budgets, explore scenario analyses, and see the results of different forecasts all in a tool that doesn’t require a data scientist to use.

Then, the marketer can share the outputs of the forecast with the finance team or other planning stakeholders so that everyone is aligned on a budget and a target that are realistic given current marketing performance.

The workflow generally looks something like this:

1. Finance team and marketing team align on high level goals and a rough budget based on last year’s media investment
2. The marketing team runs the budget through the MMM and iterates toward a budget that is both realistic (based on their knowledge of the marketing channels) and gets as close to the goal as possible
3. The marketing team brings the revised budget and target revenue back to the finance team before the process starts again

This can be a much more productive process than trying to do the whole budgeting and planning process in an excel sheet using only rules-of-thumb and heuristics that no one actually believes.

In the most advanced use case, the MMM should actually be able to recommend to the marketer an “optimal budget”. This can be tricky because the optimization needs to be appropriately constrained based on the realities of the business and building a constrained optimization tool is tricky from both a user-interface as well as an algorithmic perspective. (If you’re interested in learning more, check out how we’ve done it at Recast!)

Validating the MMM Forecasts

In order to actually use these forecasts, the whole company needs to be able to trust the forecasts that come out of the MMM. We’ve found that the best way to build up trust in the forecasts is via two techniques:

1. “Backtesting” the forecasts to show how the model “would have” made predictions historically (this is like how hedge funds test new trading strategies).
2. Demonstrating consistently accurate forecasts over time

The backtesting procedure can (and should) be done at initial model build time by repeatedly re-training the model up to certain points in time and then having the model forecast the data that it didn’t see during the training process. 

In order to consistently demonstrate accurate forecasts over time, an MMM should do the following:

1. Have an automated process for running the MMM on some consistent basis (we recommend weekly)
2. Save the model’s parameters each week in a format that can be used for forecasting
3. Repeatedly test that model’s predictions against actuals as you roll-forward in time (i.e., test the 60 day accuracy on the model trained 60 days ago)

This will allow you to demonstrate to the whole company that the model is able to consistently make accurate predictions into the future and therefore can be trusted for ongoing decision-making in the planning process.

Note that if you’re developing this in-house, there are surprising ways that this type of backtesting can go wrong. We’ve written extensively on the right ways to do backtesting in this blog post.

Conclusion

Marketing mix models are powerful tools that a business can use to get insight into marketing performance but they’re also an important tool for both short- and long-term planning that can help the business make set realistic targets and do scenario analysis on marketing budgets before committing to a plan.