Peripheral Agentic MMM
Lot of MMM vendors have started to use the word ‘Agentic MMM’. As a person who started with TF-IDF embeddings and have seen the evolution of NLP to AI, I somehow can’t come to terms with the word ‘Agentic MMM’ yet.
I believe at the current stage, it is a bit of misnomer.
In my opinion, for something to be truly Agentic MMM, the core nucleus of MMM should be agent-driven.
That nucleus is the model itself:
– Variable selection
– Functional form decisions
– Adstock specification
– Saturation curves
– Bias diagnostics
– Causal validity checks
AI still can’t do this reliably. Not at the level where you would trust it with millions in budget decisions.
📌 Where AI Actually Works Today in MMM
AI is genuinely useful in MMM. We pioneered the first MMMGPT (a RAG based model) way back in 2024 when other players just threw a bunch of documents at ChatGPT, created a wrapper and called it MMM-GPT.
However AI in MMM still works well only at periphery or Peripheral layers.
These include:
– Data cleaning & transformation
– Dashboarding & visualization
– Scenario simulations
– Model Insights summarization
– Budget allocation interfaces
But all of these are not the core modeling problem.
📌 Why the nucleus ‘The MMM model’ is still human-led
MMM is not just curve fitting or a prediction problem. It is a causal identification problem at the heart of it.
AI today Struggles with endogeneity, cannot reliably detect spurious relationships, Overfits silently when left unchecked and lacks true causal intuition
In short: It can assist thinking, It cannot replace thinking.
📌 Peripherical Agentic MMM is a Good Start
Soon we may be able to build the core MMM model with AI reliably. But the current starting point is not bad either.
Currently, Peripheral MMM helps us with increasing our throughput. It makes insights available at our finger tips or just away from a verbal utterance “eMMMy, what was the ROI of meta in 2024, compare it with 2025”.
p.S: eMMMy is our ‘almost’ all knowing MMM AI agent 😎.
All of these are made possible and we have immense trust in our AI products because – The model’s spine is still human-built. The statistical responsibility still sits with experts.
📌 We are not in the era of Agentic MMM.
We are in the era of AI assisted MMM with human accountability. Personally, I like it this way and I believe the same philosophy will resonate with lot of clients too 🤔.
