Blogs

MMM – A Collection of Tiny Experiments A lot of people see MMM and Experiments (Geo Lift / Incrementality experiments) as totally distinct things. But if you pay attention and think deeply, you will notice that MMM is not that different from Experiments. Let me explain: 📌 MMM is a Collection of Tiny Experiments. In case of an experiment: one makes a change to the variable, hold other variables constant (or at least assume it) and then observes the impact on outcome. However in real world, especially marketing we don’t get this luxury. Everything moves at the same time because multiple media / marketing spends are implemented at the same

There is no uncertainty in MMM Bayesian Analyst: We prefer Bayesian MMM because it helps us quantify uncertainty. Me: But there is no uncertainty in marketing attribution. BA: Pls elaborate. Me: Sure, you see uncertainty arises when we think the dependent variable’s value is unknowable and it behaves like a random variable. In case of MMM, we already have the sales numbers (or any KPI for that matter). They are realized and are in the past. BA: But don’t we also try to find how much each variable contributed to the KPI? Me: Yes, we do but this is the next step. First we decide which KPI we try to

Ghost MMM: Modeling Your Competitor Without Their ‘Complete’ Data For those of you who are F1 fans, you will recollect a cool analysis F1 came up with a year ago – Ghost Car Lap. This comparison was designed to analyse qualifying performance of cars, enabling analysts to overlay a transparent car representing the fastest lap over the current driver’s run to show where time is gained or lost. Couple of days ago, I was chatting with a very senior CMO and he asked me whether we model competition along with the brands’ regular MMM. Yes, sometimes we do. In domains like CPG, you can indeed build what I call a

Real AI Solution Providers Speak in Specifics Recently I came across a post from a MMM vendor benchmarking their “AI capabilities” against a few other vendors. Most of the listed capabilities were extremely surface level: – “Conversational AI” – “AI insights” – “AI reporting” – “Natural language recommendations” At some point it felt like the same capability was being stretched into 2-3 different bullet points with different wordings 🙂. Also each capability was rated differently (one was on a scale of 4, 5, some even on scale of 8. Obviously it has no statistical or mathematical meaning except to pad up the final score which was not even weighted 🤡!

Should we forget the older data in Marketing Mix Modeling (MMM)? One problem in MMM that very few discuss deeply is: How much should the past influence the present? The industry thumb rule is to use at least 2 years of monthly data or 1 year of weekly data, mainly to capture seasonality effects reliably. But people also recommend “More the better” for data. But what if the oldest year behaved very differently from the current? In year 1, lets say the brand experimented on a digital channel for the first time and experience good ROAS, low saturation effects and very high contribution. But in later years, the effect wanned

Good MMM AI Prompts Come From Good MMM Analysts Can AI build MMM models in mins? Yes. But would it be accurate and reliable? Not all. AI maturity still has a long way to get to that level of accuracy. At Aryma Labs, the core nucleus ‘The MMM model’ is still built by us (humans). But we leverage AI innovatively to generate insights, disseminate insights, perform deterministic tasks like budget optimization. Recently we demo-ed our products MMM Synapse, Nebula, Singularity and the yet to be publicly released ‘PPT add-in’ (this is gonna be real game changer) to a group of CMOs, Brand and Media managers. Stay tuned for this release. One

Meta’s MCP Server and Similar MCPs Could Create the Premier League of MMM Vendors In my last post, I talked about Marketing Measurement’s foray into MCPs and whether it is an unclear bet? Most discussions on MCP focus on the obvious benefit: AI can now talk directly to marketing platforms. But I think a bigger opportunity lies elsewhere. Here is my crazy idea 🙂 📌 The Real Opportunity: Ad Set Level Measurement Today, most MMMs operate at a relatively aggregated level – Channel level. MCP potentially opens a future where MMM vendors can access highly granular advertising data directly from platforms. Imagine building MMMs at the level of: • Ad

Marketing Measurement’s foray into MCPs – An unclear bet? Everyone is talking about MCPs in the Marketing Measurement world. Recently, several measurement vendors have also started exposing their MMM and incrementality platforms through MCP servers. It is not some massive technical breakthrough. But it does provide benefits under certain assumptions. 📌 Why the MCP play? If you have seen Breaking Bad, you know one thing: distribution is everything 🙂. In other domains too distribution is everything. Most measurement vendors are betting that usage of Claude, ChatGPT and similar AI interfaces will explode in the coming years. They are not wrong. But the big question is : – What fraction of

Is objective function optimization the right approach in Marketing Mix Modeling (MMM)? A lot of technical discussion that should be had in Marketing Measurement is never had. One such question is “Why MMM can’t be fully automated?” We already have a post on why- link in comments. When I started to think more deeply, I found another reason. 📌 Objective Function An objective function is a mathematical function that defines the goal of an optimization problem. It represents the single metric you are trying to either maximize or minimize by adjusting specific input variables. 📌 Why is this a problem for MMM? In MMM, our primary goal is to hunt for

MMM Synapse vs Copilot We recently got the opportunity to demo MMM Synapse to a seasoned MMM expert. The person was very impressed and asked a important question: “How is MMM Synapse different from Copilot or similar AI tools?” It is a fantastic question. Firstly I believe, the generation part is still not completely solved, especially the visual imagery and video aspect of it. The generation part is currently solved only for code generation and language. You can see why many companies are now pivoting to the Anthropic’s playbook. My belief is that general purpose LLMs will eventually dominate generic tasks. The real edge for startups now is going niche