The Allure (and the Trap) of Detail
The modern marketer has access to more data than ever before. The instinct to cut and slice until every campaign, channel, and audience segment is isolated feels logical. The more granular the model, the closer you get to discovering the elusive “truth” of what drives sales and ROI. But this thinking misses the fundamental truth about how effective MMM actually works.
Will Marks dissected a familiar trap: marketers new to MMM who are looking for insight by channel, product category, geography, sales channel, customer type etc, and who are told “That’s too granular”. The predictable reaction? Frustration, suspicion, and occasionally, a search for rogue modelers who promise the impossible. But no amount of statistical wizardry can bend the laws of probability, and that matters more than ever at scale.
Signal vs. Noise: The Physics of Data
The hard reality is that statistical noise is not a tech problem, it’s a law of numbers. In the most basic simulation, take a healthy business pumping out thousands of transactions per week: the data, in aggregate, is stable and insight emerges easily. But chop your data into micro-segments (by week, region, sales channel, or any other dimension) and consistent patterns evaporate. Volatility reigns.
The more granular you go, the greater the risk that random variation (pure chance) swamps genuine business signals. If sales spike in a miniature segment, marketers declare heroics. If sales drop, they panic. But nine times out of ten, it’s just natural random variation, not the work of a “world-class” creative or a doomed piece of copy. MMM algorithms, especially the Bayesian variety, are built to manage noise, but at too fine a grain, even the most robust models struggle to separate signal from chaos.
Instability: The Enemy of Model Governance
There’s another, less discussed, casualty of fine granularity: model stability. Imagine assembling a dinosaur skeleton from scattered bones. With only a few large bones, most paleontologists reach roughly the same conclusion about its form. Hand them hundreds of fragments, though, and consensus collapses. Everyone builds a different dinosaur. In MMM, this instability means that models become less reliable and more subjective as the number of variables and cuts skyrockets.
Without rigorous model governance and open-source testing, it’s almost impossible to flag when a finely sliced model is producing unreliable, even fictional, results. Marketers may think they’re gaining insight, but they’re actually getting plausible-sounding fiction.
Granularity That Matters: Align with Decision-Making
The sweet spot for MMM lies not in maximum detail but in the level of granularity that matches how decisions are actually made. Marks advocates for using MMM to answer big, consequential questions: which channels to prioritise, how to allocate budget to hit revenue targets, and how to balance brand and performance investment.
Instead of chasing every possible cut for the sake of reporting, marketers should focus on insight that drives action. The temptation to detail every variable (channel, creative, region, segment) should be weighed against one hard question: Will this split help me make a decision that drives meaningful business value?
Inputs: Less is More for Accuracy
This isn’t just about outputs. Overloading the model with “everything” (every piece of data, every campaign, every variant) actually reduces accuracy. Selective, high-quality inputs let the model identify and attribute value more reliably. Only after the right structure is in place does deeper analysis unlock true, actionable nuance. Model first for the big levers. This unlocks the ability to go deeper and dive into the role of more granular drivers.
Lessons from MMMorning!
Will Marks’ core message is one the industry needs to internalise: granularity for its own sake doesn’t unlock value. The job of MMM is to reflect reality. Not fabricate a comforting fiction that tells every stakeholder exactly what they want to hear. Pursuing goldilocks granularity, or a level that is “just right” for stability, reliability, and genuine action, creates models that decision-makers can trust.
For marketers staring down dashboards jammed with micro-insights, the message is clear: resist the urge for ever-finer detail. Insist on stable, testable, and decision-relevant models. Only then does MMM deliver what matters most: a platform for action, not just a mirror for today’s complexity.
Watch the MMMorning! episode here.
[Will Marks is Head of Marketing Science at Mutinex]