The setup
Most marketing measurement is one tool stretched to do three jobs it wasn't built for. Platform attribution is asked to prove incrementality. Last-click is asked to allocate budget. Spreadsheet MMM is asked to call campaign-level shots. None of those work, and a CFO with a spend question deserves better than a deck full of platform-reported credit.
The PCG Way is three tools, used together, each scoped to the question it can actually answer.
Incrementality Experiments: what was caused
If a channel went dark tomorrow, what would actually be lost? That's the question incrementality answers, and nothing else does.
We run geo holdouts, audience splits, ghost-ad tests, and platform-native lift studies depending on what the data and the program will support. The point isn't sophistication; the point is causal lift. A clean test on a real-world hypothesis beats a beautifully modeled correlation every time.
Incrementality is the strongest signal in the toolkit. It's also the slowest and most expensive to run. You can't test everything, so you test the calls that matter: the channels carrying real spend, the audiences you're not sure are working, the structural choices the platform won't tell the truth about.
Multi-Touch Attribution: who did what along the way
Incrementality tells you if a channel moved the needle. MTA tells you how. Which touchpoints showed up in winning paths, which were dead weight, where the platform's default credit is wrong.
We work with two layers. The first is platform-native: Google Ads conversions, GA4 path data, the standard model comparisons. The second is a Bayesian model we run when the program is big enough to support it, calibrated against incrementality tests so the credit it assigns has a real-world anchor.
MTA on its own is dangerous. It will confidently allocate credit to whichever channel sits late in the funnel, regardless of whether that channel caused anything. Calibrated against incrementality, it becomes useful: a tactical lens for which assets, audiences, and keywords are doing the work inside the channels that already passed the causal test.
Marketing Mix Modeling: how to allocate the next dollar
Incrementality is the per-channel truth check. MTA is the per-touchpoint diagnostic. MMM is the portfolio view: given everything we know about media, seasonality, pricing, distribution, and macro drivers, where should the next dollar go?
MMM is the only tool of the three that can speak in dollars at the channel-portfolio level. It's also the only one that captures brand effects, long-lagged response curves, and saturation. For programs above roughly $5M/year in media, it's the layer that holds the C-suite conversation together.
MMM has known weaknesses: it needs two years of clean history, it's slow to react to platform changes, and the output is only as good as the inputs and the analyst. Run alongside incrementality and MTA, those weaknesses are bounded. Run alone, they're the whole story.
Why all three
Each tool has a job that the other two can't do well:
- Incrementality = causation
- MTA = mechanism
- MMM = allocation
A program that only runs platform attribution will keep over-crediting the channels that show up last. A program that only runs MMM will miss the platform-level changes that show up week to week. A program that only runs incrementality will run out of testing budget before it answers the next question.
The PCG Way is what it looks like to use the three together: each scoped to its strength, each calibrated against the others, each in service of a decision a CFO will sign.
The work, in order
When a new program starts, the order is roughly:
1. Stand up clean conversion architecture (server-side, enhanced conversions, value flows)
2. Get the platform-native MTA reading correctly
3. Run the first incrementality test on the channel carrying the most spend
4. Use that test to calibrate the platform credit you trust
5. Add MMM once you have enough history and spend to support it
The reason it's an order, not a checklist, is that step 5 doesn't work if steps 1-4 haven't run. MMM trained on broken inputs is worse than no MMM. The discipline is upstream.