Marketing mix modelling has re-emerged as the most credible methodology for understanding marketing's commercial contribution. But the methodology's rehabilitation has introduced a new risk: treating its outputs as objectively true ROI figures rather than statistical estimates with real limitations.
Marketing Mix Modelling as a Strategic Instrument
Marketing mix modelling has undergone a significant rehabilitation in the past five years. Once dismissed as a relic of pre-digital measurement — too slow, too aggregate, too expensive for the always-on performance environment — MMM has re-emerged as the most credible methodology available for understanding the true contribution of marketing investment to business outcomes. The driver of this rehabilitation is not nostalgia. It is the accelerating inadequacy of digital attribution in a world where cookies are disappearing, consent requirements are tightening, and platform-reported figures have become demonstrably unreliable.
But the rehabilitation of MMM has come with a new risk: the conflation of what the methodology can genuinely deliver with what organisations hope it will deliver. Marketing mix modelling is a powerful instrument when properly constructed and correctly interpreted. It is a significant liability when treated as a black box that produces objectively true ROI figures, or when its inherent limitations are not understood by the executives acting on its outputs.
Understanding both the genuine capabilities and the hard limitations of MMM is now a prerequisite for any senior marketing leader engaged in budget defence or capital allocation. The methodology is not optional — it is increasingly the standard of evidence that boards and CFOs expect. But the standard of evidence is only valuable if the people receiving it understand what it actually represents.
What MMM Genuinely Reveals
At its core, marketing mix modelling uses statistical regression techniques — increasingly augmented by Bayesian methods and machine learning — to decompose observed business outcomes (typically revenue or volume) into their contributing factors. These factors include marketing activities across channels, baseline sales driven by brand equity and seasonality, pricing and promotional effects, distribution changes, and macroeconomic conditions. When properly specified, the model isolates the incremental contribution of each marketing investment to the observed outcome.
The genuine insights MMM produces are substantial. It provides cross-channel comparability on a common outcome metric — allowing paid search, television, out-of-home, and social media to be evaluated against the same revenue contribution standard. It captures both short-term activation effects and, when configured with sufficient historical data, the longer-term brand effects that digital attribution systematically ignores. It is immune to the double-counting problem that afflicts multi-touch attribution, because it works from observed aggregate outcomes rather than individual user journeys.
Marketing mix modelling is powerful when correctly interpreted. It is a liability when treated as a black box producing objectively true ROI figures.
MMM also provides something that digital attribution cannot: an estimate of the base sales contribution — the revenue that would occur even without any marketing activity. This figure is strategically significant. Organisations that discover their marketing is contributing 15 per cent of revenue to a base of 85 per cent have a fundamentally different budget conversation than those whose marketing drives 60 per cent of incremental sales. The model makes this visible for the first time in many cases.
The Hard Limitations That Practitioners Rarely Advertise
MMM has three categories of genuine limitation that are frequently understated in vendor presentations and agency recommendations. The first is data requirements. A credible MMM requires a minimum of two to three years of weekly data across all modelled variables — marketing spend, outcomes, pricing, competitive activity, and external factors. Organisations with gaps in historical data, recent structural business changes, or limited market data availability will produce models with wide confidence intervals that are effectively unusable for precise budget allocation.
The Interpretation Gap Between Analysts and Executives
The most consequential limitation of MMM is not technical. It is interpretive. The gap between what a well-constructed model actually says and what executives take from it in budget discussions is frequently enormous — and the consequences of that gap are significant.
MMM outputs are statistical estimates with confidence intervals, not measurements. The headline ROI figure for a given channel is the central estimate of a distribution. Depending on model quality and data availability, the true value could be meaningfully higher or lower. Budget decisions made on point estimates without acknowledging this uncertainty are not rigorous — they are applying a veneer of quantitative rigour to what remains a judgement call.
Furthermore, MMM measures average return on historical spend levels. The relationship between spend and return is typically non-linear — there are diminishing marginal returns at higher spend levels, and sometimes increasing returns at low spend levels due to minimum effective frequency thresholds. An ROI figure derived from last year’s spend level may not be a reliable guide to the return on an incremental dollar at a different spend level. This distinction matters enormously for budget optimisation and is frequently absent from executive presentations of MMM results.
Using MMM as One Input Among Several
The most sophisticated marketing organisations treat MMM as one input in a triangulated measurement system rather than the definitive arbiter of budget allocation. MMM provides strategic-level channel contribution estimates. Incrementality testing — controlled experiments that isolate causal effects at a tactical level — validates and refines the channel-level findings. Brand tracking research provides the leading indicators of long-run demand that MMM alone cannot capture in real time. Together, these methodologies produce a measurement system that is genuinely fit for board-level decision-making.
For Australian marketing leaders, the practical implication is twofold. First, commissioning MMM without investing in the interpretive capability to use its outputs correctly is not an improvement over the status quo — it is a more expensive version of the same problem. Second, the value of MMM is maximised when it is refreshed regularly — ideally quarterly or biannually — rather than treated as a once-per-year strategic exercise. The commercial environment changes continuously. Measurement frameworks that cannot keep pace with that change are informing yesterday’s decisions with yesterday’s data.