Media Mix Modeling (MMM)

MMM works by building a regression model that correlates changes in channel spend with changes in revenue over time, controlling for external variables like seasonality, pricing, and competitor activity. The model estimates the marginal contribution of each dollar spent in each channel — not by tracking individual users, but by observing the relationship between spend patterns and revenue outcomes at the aggregate level.

Traditional MMM required large datasets (2+ years of weekly data) and significant statistical expertise, and was primarily used by large CPG and retail brands. Lightweight MMM tools — including Meta's Robyn and Google's Meridian, both open source — have made the technique accessible to mid-market brands. Results are directional rather than precise, but calibrated with incrementality data, they provide a reliable picture of channel contribution.

Exactius uses MMM as a complement to incrementality testing — not a replacement. Where incrementality testing provides precise causal measurement for specific channels, MMM provides portfolio-level visibility across all channels simultaneously, including those where individual-level testing is not feasible.

MMM answers the measurement question that user-level attribution cannot: what is the contribution of channels with no direct click path — TV, radio, out-of-home, podcast, brand sponsorships? These channels move the market but leave no trackable footprint in attribution models. MMM detects their contribution through the revenue signal they create.

Since iOS 14 degraded user-level signal, MMM has become a more important input for digital channel measurement as well. When platform-reported ROAS is partially modeled, an MMM that uses actual revenue as its dependent variable provides an independent cross-check on whether those channels are driving the outcomes platforms claim.

For executives allocating capital across a media portfolio, MMM answers the highest-level question: given everything we spent last quarter, where did the revenue actually come from? It surfaces the channels that are invisibly driving demand and the ones that are consuming budget without moving the needle.

What MMM requires

MMM requires at least 52 weeks of weekly data (2 years is better for seasonality detection): total revenue by week, spend by channel by week, and external variables (holidays, price changes, competitor activity). The more channels included, the more precisely the model can isolate each channel's contribution.

Calibration with incrementality

MMM outputs are most reliable when calibrated with incrementality test results. If an incrementality test shows Meta driving 30% incremental lift, and the MMM attributes 45% of revenue to Meta, the model is over-attributing and needs recalibration. Running at least one geo holdout per major channel per year gives the calibration anchors the MMM needs to stay accurate.

Tools: Meta Robyn (open source, R-based), Google Meridian (open source, Python), and commercial tools like Northbeam and Triple Whale offer MMM-adjacent measurement. For most mid-market DTC brands, Northbeam or a custom lightweight model is sufficient.

MMM experienced a justified revival after iOS 14 because it exposed what sophisticated measurement teams already knew: user-level attribution was always an approximation, and aggregate measurement often told a more honest story. The difference is that before ATT, the approximation was good enough to act on. After ATT, it frequently is not.

Exactius uses lightweight MMM as a portfolio-level sense check on the Capital Allocation Loop's channel-level decisions. When MMM attribution diverges significantly from incrementality-adjusted platform data, it flags a measurement gap that requires investigation before budget is moved.

David Manela's position is that MMM and incrementality testing are complementary — MMM for the full-portfolio picture, incrementality for channel-specific causal confirmation. Neither is sufficient alone: MMM without incrementality calibration over-attributes high-spend channels; incrementality without MMM misses unmeasured channels entirely.

Exactius embeds growth squads that build and maintain MMM infrastructure as part of the measurement stack for partners spending above $500k annually on media. Below that threshold, MER triangulation and incrementality testing typically provide sufficient signal without the complexity of a full model.