What MMM Is (and Isn't)
Track 0 - MMM in 60 Minutes · Module 0.1
PM / everyone
Course home
Learning ObjectivesModule 0.1 · ~15 min
State the four client questions Rapid ROI answers.
Distinguish MMM from multi-touch attribution (MTA) and from incrementality testing, and say when each is the right tool.
Explain why aggregate, regression-based measurement exists at all: it is privacy-safe and works without user-level tracking.
The four questions the client is paying for

Strip away the math and Rapid ROI answers four business questions, straight from the program's own pre-read:

  1. Allocation, forward-looking: how much should I spend on each channel, format, and platform next cycle to maximise sales and brand power?
  2. Measurement, backward-looking: what sales did each media channel, platform, and campaign actually generate, and at what ROI?
  3. Decomposition: what share of my total sales is driven by each of the 4P levers - and what is actually contributing to my growth?
  4. Time horizon: how does my media drive both short-term sales and long-term brand strength?

Everything else in this course - adstock curves, bound setting, Kalman filters - exists only to make those four answers trustworthy.

MMM vs MTA vs incrementality tests

New joiners from digital-marketing backgrounds usually arrive knowing one of the other two measurement families. The differences matter because clients mix them up constantly:

ApproachWhat it doesGrainStrengthsBlind spots
MMM (this program)
Regression on aggregate weekly data: sales vs media, price, promo, distribution, macroMarket / brand / weekCovers ALL channels incl. offline; privacy-safe; captures price and promo too; long historyCannot see individual users or creative-level nuance without extra data; needs 2-3 years of history
MTA (multi-touch attribution)
Distributes credit for a conversion across the user's tracked touchpointsUser / impressionVery granular; near real-timeOnly works where users are trackable - broken by cookie deprecation and walled gardens; blind to TV, OOH, price
Incrementality test
Controlled experiment: expose one region/group, hold out another, measure the differenceTest cellCleanest causal read availableExpensive, slow, one channel at a time; you cannot experiment on everything

The three are complements, not rivals. A mature advertiser runs MMM as the always-on portfolio view, uses experiments to calibrate or sanity-check specific channels, and keeps MTA (where it still works) for in-flight digital tactics. The KT sessions frame incrementality tests as the "chew test" comparison point: a direct bite of causal evidence for one channel, versus MMM's full-menu view of every lever at once.

Why aggregate regression at all? Because it needs no user-level tracking. MMM reads weekly totals - sales, GRPs, impressions, spend - so it is unaffected by cookie loss, app-tracking opt-outs, and privacy regulation, and it can treat TV, print, and a price change with exactly the same machinery as digital.

Myths vs reality
  • "MMM will tell me which ad creative worked." Not natively - the standard grain stops at campaign level (L3). Creative-level reads exist only in the long-term model's L4 layer, and only where the data platform carries creative metadata.
  • "MMM is just correlation." It is constrained regression with business priors: sign constraints, bounded coefficient ranges, and transformation theory (adstock, saturation) are all there to force the fit toward causally plausible answers. Track 1.6-1.7 is entirely about this discipline.
  • "Digital is measurable, so MMM is for TV." Platform-reported digital numbers measure attribution inside that platform's walled garden, generously. MMM referees across platforms on one consistent yardstick.
Check Yourself
A client wants to know whether their branded-search spend is truly incremental, and they are willing to wait a quarter for the answer. Best tool?
Why: one channel, a causal question, and time to run a controlled experiment - that is exactly the incrementality test's sweet spot. MMM would give a modelled answer, and branded search is notoriously confounded with existing demand in both MMM and MTA.
Cookie deprecation just broke the client's attribution vendor. Why is the MMM unaffected?
Why: MMM's inputs are weekly totals (spend, GRPs, impressions, sales). Nothing user-level enters the pipeline, which is precisely why the industry re-embraced MMM as tracking degraded.
Which of the four Rapid ROI client questions is forward-looking?
Why: allocation is the forward-looking question, and it runs on marginal ROI (mROI) and response curves rather than historical ROI - a distinction module 0.3 and Track 3.3 develop fully.
Sources
Authored from:
  • UL_Rapid ROI_Pre-Read_Document 1.pptx slides 2-4 (the four business questions, what Rapid RoI provides, how it works)
  • UL - KT (1).docx (4 Jun session: MTA / incrementality / chew-test contrast)
  • MathCo Methodology Understanding_UL.xlsx sheet MMX Context
  • External framing: Jin, Wang, Sun, Chan & Koehler, "Bayesian Methods for Media Mix Modeling with Carryover and Shape Effects" (Google Research, 2017) - the shared academic ancestor of modern MMM tooling; cited for industry context even though this program is not Bayesian (see module 3.5)