9 points a marketer should know about marketing mix modelling

1. The closest analogy of MMM is randomised trials embedded into every campaign. If randomised controlled trials (RCTs) are the gold standard of marketing measurement, MMM is a poor man’s RCT with less costs and disruptions.
2. It is called “modelling” as opposed to “calculation”, because:
   1. it relies on assumptions of how the overall result depends on various factors,
   2. the algorithm seeks parameter values for those dependencies so that the model best explains the actual results.
3. Acronym MMM could stand for marketing mix modelling or media mix modelling, the former being more generic and taking into account not only media, but also external factors, pricing, etc.
4. MMM works where there is no direct tracking of individuals: for TV / Radio, for digital and social displays, for multi-channel marketing. But it also can be used to detect and quantify all kinds of dependencies, e.g., is there a material link between the “likes” to your social media post and sales?
5. Technically, MMM still measures coincidence rather than causality, as both the results and the inputs could be linked to another factor, e.g., both sales and advertising can be linked to events / holidays.
6. Models can be predictive, i.e., of such high quality that you can use them to forecast future results and/or plan, or they can be explanatory / directional – simply explaining historical results and suggesting which segments of your campaigns should be given a boost, and where to slow down. In the latter case, you need less data, and can optimise your campaigns in near real time.
7. There is no upfront accuracy guarantee. Instead, as you construct a model, listing various factors that can affect the result, the algorithm will calculate how accurately that specific model is. If the accuracy is unsatisfactory, you could be missing a factor, or the model assumptions could be wrong, or the results were influenced by once-off event. Add, adjust, and retry.
8. MMM has 4 key limitations:
   1. It requires loads of well prepared and pre-classified data, so that effects are matched to their potential causes.
   2. It is necessarily holistic; you cannot just focus on, say digital advertising, completely ignoring TV. If you do, the model will be inaccurate (see point 7)
   3. It requires that the inputs vary somewhat in the course of campaign.
   4. It cannot untangle influence of correlated factors, e.g., if you always use TV and Radio together, the algorithm cannot report on their effectiveness separately.
   You need to take these limitations into account when planning your campaign in order to measure confidently.
9. It can only measure short-term effects of marketing activity, typically a few weeks after that activity. On its own, without any additional techniques, it cannot measure long sales cycles.

Overall, MMM is not a magic bullet that would just give you all the answers in a good-looking dashboard. But it gives you an end-to-end top-line view, provides key benchmarks, facilitates optimising campaigns, and helps understanding what drives your marketing success.