The Inference-to-Description Ratio
We just got a pat on the back from one of our clients, who said “Your MMM deck have a lot more inference than our previous vendor”.
This made our day, we have always strived to provide more insights than superficial description of data.
đź“Ś I call this the Inference-to-Description Ratio.
I believe good MMM Vendors have a high Inference-to-Description Ratio.
Basically, it is how much of the work is Inference vs Description
Most MMM decks (especially in EDA stages) are full of descriptions:
“Sales peak in December”
“TV spend increased in Q3”
“Digital contributes 35% of spend”
These are not insights. These are mere narration. Even a dashboard can do that or any LLMs for that matter.
However, I think most clients ask more:
– Why did it happen?
– What does it imply for decision making?
– What should we do next?
So if some young analyst is reading this, perhaps instead of saying
“Sales peak in December”, one could add more insights by saying:
“December peaks seems to be driven by promo intensity + retail expansion, not just seasonality. MMM model to confirm”.
I think the latter adds more insight even if it is just directional and not confirmed by MMM yet.
đź“Ś How to Infer Better?
I believe one cannot infer without context and deep business understanding.
And context doesn’t come from R-squared value, p-values or Fancy priors
It comes from understanding:
– Domain that business operates in
– The market share
– Pricing strategy
– Distribution constraints
– Trade vs media interplay
– Category dynamics
– Consumer behavior
Without all this, MMM becomes a statistical storytelling exercise.
At Aryma Labs, we have consciously built a different kind of team.
We are not just a group of geeky statisticians optimizing loss functions.
We think in terms of:
ROAS, MROAS, CLTV, MER, churn, distribution, brand equity (TOMA, spontaneous recall), reach, frequency, brand vs performance; as comfortably as we think in adstock, MAPE and parameter stability.
Because ultimately, MMM is not built just to explain the past or just predict the future, It is built to influence the future.
And if your vendor cannot translate coefficients into business consequences, you don’t have an MMM partner. You have a reporting tool !!
