MMM: Do your models include long-term impact?
Are the models robust?
Most MMM models focus primarily on short-term sales uplift often measuring impact within a few weeks or months of activity.
But many channels, particularly brand-building ones, influence behaviour long after a campaign ends. They build memory structures, increase mental availability, and shape future demand. If long-term effects aren’t captured properly, the model risks undervaluing channels that create sustained growth rather than immediate spikes.
There are credible ways to account for this. Some models incorporate brand health metrics, such as awareness, consideration or preference, linking media activity to brand shifts and then to future sales. Others use dedicated long-term modelling frameworks to estimate how brand equity accumulates and decays over time.
Some models take a shortcut by applying generic “long-term multipliers” to short-term ROI. While this can inflate headline numbers, it doesn’t necessarily improve understanding or guide better optimisation decisions as there is a risk of over-simplifying the approach.
A well-constructed approach is likely to use evidence-based and specific category insight to consider the long-term impact.
For Out of Home advertising, which can drive both immediate response and longer-term brand effects, this distinction is critical. Long-term value should be grounded in evidence and not assumption.
How can I check if long-term channel effects are included in MMM?
- Ask – are brand metrics are included? Are awareness, consideration or preference modelled as intermediate steps between media and sales?
- Understand the methodology. Was a recognised long-term modelling framework used, or was a simple multiplier applied?
- Check the evidence base. Are long-term effects derived from real data and brand tracking, or borrowed assumptions?
- Test sensitivity. Do long-term ROI figures remain stable under alternative assumptions — or do they swing dramatically?
If long-term impact changes significantly depending on arbitrary inputs, treat it with caution.
Do my confidence intervals and t-test stats show that the results are reliable?
YES!
Results show strong statistical confidence.
Good - evidence supports decision-making.
Gain more confidence by sense-checking signals for an intuitive sense of each channel based on your knowledge of campaign performance, check that the confidence interval is narrow enough and verify t-stat robustness that shows performance across several campaigns.
NO!
Confidence is low.
ROI figures may vary widely campaign to campaign where you have a large confidence interval.
Be sure to treat figures as indicative and with caution.
Look to strengthen inputs and validate the outputs with supporting evidence.
To understand more: