A measurement stack audit before Q4

Q4 volume exposes gaps that Q2 volume hides. A postback delay that costs a team a few hours of attribution clarity in a quiet month becomes a live budget-allocation problem when daily spend triples. So before the quarter ramps, treat the measurement stack itself as something to test rather than something to assume still works the way it did in January.

Confirm what actually changed underneath you

Several things moved this year that touch measurement even when nobody framed them as measurement news. AppsFlyer closed a Series E in June at a $2.7 billion valuation; Google and Meta both took minority stakes, and so did Unity and Moloco, which leaves a meaningfully different ownership structure from the one your integration assumed. Adjust's device-level migration of Tripledot's entire live portfolio in March, moving billions of install records without disrupting live campaigns, is a reminder that vendor-side infrastructure work happens continuously and mostly invisibly. Neither change requires action on its own. Both are reasons to re-confirm, rather than assume, that vendor documentation matches what your account is actually doing today.

Check postback logic against SKAN 4 and AdAttributionKit, not last year's notes

WWDC in June left AdAttributionKit and SKAdNetwork untouched, which is good news for stability and bad news for anyone putting off a proper audit because "nothing changed." Nothing changing is exactly when things drift. Postback windows slip out of sync with what the dashboard assumes; so do conversion value schemas and coarse-value fallbacks, because nobody revisits a pipeline that appears to be working. Walk the full chain again: conversion value definitions, timer windows, what happens on a coarse-only postback, and whether your reporting layer flags coarse conversions distinctly from fine ones rather than blending them into a single number.

Re-run your incrementality baseline before spend ramps

Adjust's March report on 2025 session data found the paid-to-organic install ratio up 61% year on year, with ATT opt-in edging up only slightly, from 38% to 39%. Read those together. A rising paid share against a flat opt-in rate means more of your install base sits in the coarse, low-signal bucket than the one your incrementality model calibrated against. If your last incrementality read predates that shift, run it again with current data before committing Q4 budget to a mix that assumes an older ratio.

Stress-test the dashboard, not just the pipeline

A stack that reports correctly at normal volume can still misreport under load, whether through rate limits or batching delays or a reporting layer that silently drops late-arriving conversions during a spend spike. Pull a sample of last year's highest-volume day and check whether every install from that day still reconciles against source data today. Any gap found now, in a quiet week, is a cheap fix. The same gap in the third week of November is a live decision made on bad numbers.

Write down what the audit found, not only what it fixed

An audit that lives only in a chat thread doesn't help the team in November, when the person who ran it may have their head down in day-to-day optimisation. Keep a short written record: every check performed, what passed, what needed a fix, and who owns the fix. That record becomes the first place to look when a number looks wrong during peak weeks, rather than starting the investigation from zero under time pressure, and it gives the following quarter's audit a baseline to compare against so drift gets caught before it compounds into another full re-check.

None of this is glamorous work. None of it produces a chart worth presenting. It is the work that decides whether the charts produced in December are worth trusting.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• A Year-End Measurement Stack Review Before You Rebudget
• The MMP Migration Checklist to Run Before You Switch