Marketing Mix Modelling, or MMM, to give it its popular acronym, doesn’t give definitive, final results – it gives actionable insights.
But what should those actions be?
Kinase Analytics Engineer Bhuvan Setty argues that things become clearer when we think of MMM as ‘a decision engine’.
MMM is a sophisticated, powerful methodology – but when the latest slides and dashboards from their MMM project arrive, some marketers end up with more questions than answers.
How do MMM projects end up going down this blind alley – and how can we map out a better way? To start with, let’s look at the problem.
What goes wrong?
MMM works by taking historical data about your marketing channels. Analysis then measures the contribution of channels and their influence on each other. Marginal ROIs are modelled allowing the marketing mix modelling analysis to output an optimised media plan to maximise for a business KPI or efficiency.
It’s a one off jolt to optimisation, and a refresh of reporting to show a ‘true’ measure extracted from the numbers being tracked and packaged up from daily attribution.
The question then is whether to action the ‘ideal budget plan’ which MMM has outlined – if it diverges from observable attribution, this may be an uphill struggle, and a compromise or even no action may result.
This familiar track is founded on two common goals set for MMM – to determine the real ROI of marketing channels, and to determine the best channel investment mix. Both would be great – but what MMM feeds back can end up unusable, with the marketing team no closer to being able to scale investment in a way which drives the business’ KPIs.
Even if a new budget mix is adopted, if MMM is seen as a one-off project, then the new investment will be a bet as the business waits to see if it is driving the promised results.
As new dynamics enter the market, the validity of the last MMM will start to be (rightly) questioned, as it drifts from the present reality.
Restoring MMM as a decision engine
MMM needs to guide investment levels and further testing. Different kinds of tests and modelling should feed into each other, better informing the marketing team over time even as the terrain changes.
What does that look like in practice? The latest cuts from your experimentation programme should feed into MMM from the start, as priors (historical and relevant data which should inform MMM). For example, an incrementality test on paid search may have run a ‘go dark’ region in order to observe the uplift paid search advertising drives.
The results from this experiment would then be fed into MMM as a prior finding for its analysis.
The results of MMM then guide and refine experiments and tests in the next period. The results will guide new incrementality tests to reduce the risks and increase business confidence in the channel mix model which MMM suggests.
This altered mindset sees MMM as a way of making and guiding decisions, rather than as an answer which is hard to implement or action. Every decision has analytical consideration.
Every input is statistically justifiable. Everything is being executed with more confidence. Importantly, the whole approach is platform independent. Hence, your results are not affected by attribution issues which bedevil digital marketing.
Moreover, now you actually can use the results to interpret your platform and overall ROIs, but this time with a robust ongoing framework behind them.
It’s the beginning of MMM as a decision engine in practice.