Every organisation approaching Marketing Mix Modelling (MMM) for the first time gets bombarded with the same pitch, dressed up a dozen different ways: a comprehensive programme, a unified data warehouse pulling in every channel and geography you might ever want to model, a multi-year commitment before a single result has been produced. It is in the vendor’s interest to propose this. It is in the consultant’s interest too – a bigger scope is a bigger fee and a longer engagement. None of that makes it the right way to start.
The honest answer, as with most things worth doing properly, is to start small and get bigger. The worst outcome I see is an organisation commissioning a data infrastructure built for every question it might one day want to ask, before it has established which questions Marketing Mix Modelling can actually answer for its business, with its data, on its timeline. That warehouse often turns out to be the wrong shape entirely – built around channels that don’t matter, missing the covariates that do, six months late and already needing to be rebuilt.
Facebook’s early engineering motto was “move fast and break things”. It got the company a long way before it stopped being true – once the platform mattered enough that stability became the priority, the motto quietly changed to “move fast with stable infra”. The order matters. You earn the right to build something permanent by first finding out, cheaply and quickly, what is actually worth making permanent. The best MMM programmes I have been involved in follow the same order: start with one focused piece of analysis, see what it does for the organisation, and only then decide what deserves the comprehensive build.
A single, well-scoped first analysis does two jobs at once. It answers a real business question your marketing leadership actually has right now, which is worth doing on its own. But it also tells you, honestly, which of your other questions Marketing Mix Modelling can address, and which ones need a different method entirely – a test, an attribution read, a piece of qualitative work. You find this out for a fraction of the cost of a full programme, and you find it out before you have committed the budget and the credibility to a comprehensive build that may be answering the wrong questions.
Ask an MMM analyst directly, “can your technique answer this question?”, and pay attention to how they answer. Nine times out of ten, if the question is genuinely relevant to marketing, the honest response is “maybe” – or, if they are being more useful, “yes, but”: yes, conditional on the data you actually hold, the timeline you’re working to, and two or three other factors they should be able to name specifically. An analyst who answers with an unqualified, confident yes on the first call, before they have seen your data, is telling you more about their sales process than about your model.
None of this is an argument against ambition. Most organisations that start small do get bigger – a properly scoped first analysis nearly always earns the case for the next one, and the one after that, because by then you know what you’re building and why. It is an argument against building the comprehensive version first, on the strength of a pitch, before anyone has proven what the comprehensive version is even for.
For the fuller picture of what a first engagement needs to get right beyond data, see The four disciplines of Marketing Mix Modelling.
If you are weighing up a first MMM engagement and the proposal in front of you starts with a data warehouse rather than a question, drop a line to hello@themmmdoctor.com.