---
title: The measurement stack — last-click attribution, incrementality and MMM together
slug: measurement-stack-attribution-incrementality-mmm
summary: Why app advertisers need both deterministic last-click attribution (for platform optimisation and daily decisions) and incrementality testing (for budget allocation), where MMM fits, and how to reconcile them against top-line results.
domain: marketing
tags: [measurement, attribution, incrementality, mmm, mmp, ad-fraud]
related: [when-last-click-is-good-enough, last-click-attribution-biases, incrementality-test-design, mmm-evaluation-and-open-source, mmp-role-and-skadnetwork, signal-engineering-and-conversion-event-audit, apple-ads-attribution-without-mmp, meta-ads-manager-metrics-setup, paid-vs-organic-cannibalization, demand-creation-vs-demand-capture-channels]
sources: [yt:iD_YbEW0Y3w, yt:qZj6_mJGOMg, yt:tJzzjP29f-k, yt:3Uj6p1N8VJA]
updated: 2026-10-24
---

# The measurement stack — last-click, incrementality and MMM together

## TL;DR
- No single method is right; stack them. Last-click (user-level, deterministic) feeds the ad platforms and daily decisions; incrementality tests decide budget allocation; MMM comes last, if at all.
- Incrementality does **not** replace last-click: Meta, Google, AppLovin and TikTok optimise on user-level conversions, which incrementality can't supply.
- Media buyers need **daily signal and ad-level granularity**. The fix is *incrementality-adjusted attribution*: calibrate last-click numbers with test results, keep the daily ad-level dashboard.
- Check every belief against the top line: if your channel numbers keep missing the forecast, the numbers are wrong.
- Ad fraud mostly shows up as non-incrementality. You rarely need a dedicated fraud-detection vendor.

## Details
**Why both.** [[people/thomas-petit]] answers "is incrementality the way?" with "yes and no". CPI, CPA and ROAS all depend on the attribution method, and incrementality is the right lens for allocating budget. But it is not user-level, and networks still need deterministic last-click events "for the next impression to show up". You form a better opinion by piling several imperfect methods together.[^1] [[people/eric-seufert]] thinks the probabilistic approaches were oversold around 2021, right after ATT.[^2] Haus, a geo-testing vendor, requires every app client to have an MMP, because incrementality reconciles against attribution data rather than replacing it.[^3]

**What practitioners actually consume.** [[people/olivia-kory]] (Haus) says the two non-negotiables are daily signal and ad-level granularity. App media buyers are "in Meta ads all day", so a month-end incrementality readout doesn't fit how they work. Haus's answer is to push experiment results back into the familiar dashboard as bias adjustments on attribution.[^4] Attribution itself remains table stakes; measurement has simply grown beyond it.[^5]

**A sensible order of build-out** ([[people/michael-kaminsky]]): touch-based attribution, then a post-checkout "how did you hear about us" survey, then some incrementality testing. Add an MMM only when you can't test fast enough to keep measuring a channel, or you need one model for forecasting and planning.[^6]

**Make your beliefs consistent.** Plug your channel CPAs into a forecast. If you keep missing it, something in the numbers is wrong.[^7] AI tools now let marketing leads ask questions like "we raised Facebook spend 30%, why didn't sales go up?" directly against the data.[^8]

**Scale changes the job.** Once spend reaches tens of millions a month across many channels, last-click "will only take you so far". You need incrementality, MMM and marginal-cost thinking: does adding a channel actually add customers?[^9]

**Fraud.** Petit's position: if a channel is largely incremental, 20% fraud inside it still leaves it incremental. Mostly-fraudulent channels show up as non-incremental in geo tests, so fighting fraud technique by technique costs more effort than money.[^10] Seufert rejects vendors who say media buyers can't detect fraud themselves, and calls a quoted $500K/year fraud-consulting fee absurd.[^11]

## Disagreements / caveats
- Kory (incrementality vendor), Kaminsky (MMM vendor) and the podcast's sponsors (INCRMNTAL, Branch) all have a commercial stake in "you need more than last-click".
- For small advertisers on one or two channels, see [[when-last-click-is-good-enough]]: the full stack can be overkill. [[apple-ads-attribution-without-mmp]] shows a no-MMP setup for Apple Ads + ASO.

## See also
- [[last-click-attribution-biases]] — which channels last-click flatters or punishes
- [[incrementality-test-design]] — how to run the tests
- [[mmm-evaluation-and-open-source]] — if you go to MMM

[^1]: yt:iD_YbEW0Y3w @ 03:24 — Thomas Petit
[^2]: yt:iD_YbEW0Y3w @ 04:10 — Eric Seufert
[^3]: yt:iD_YbEW0Y3w @ 07:29 — Eric Seufert (relaying Olivia Kory)
[^4]: yt:qZj6_mJGOMg @ 23:05 — Olivia Kory
[^5]: yt:qZj6_mJGOMg @ 11:38 — Eric Seufert
[^6]: yt:tJzzjP29f-k @ 43:21 — Michael Kaminsky
[^7]: yt:tJzzjP29f-k @ 46:40 — Michael Kaminsky
[^8]: yt:tJzzjP29f-k @ 38:34 — Michael Kaminsky
[^9]: yt:3Uj6p1N8VJA @ 44:32 — Nebojsa Radovic
[^10]: yt:iD_YbEW0Y3w @ 12:11 — Thomas Petit
[^11]: yt:iD_YbEW0Y3w @ 12:11 — Eric Seufert