Attribution windows: the lookback debate

Meta's default attribution window credits installs and events for up to seven days after a click and one day after a view. Google Ads counts conversions differently by campaign type, and for some conversion actions it lets an advertiser choose a window as long as ninety days. TikTok's default sits at seven days post-click and one day post-view, which matches Meta's headline numbers while diverging in how it treats a view-through touch when the user later converts through another channel entirely. Each MMP then applies its own deduplication logic on top of all that, deciding which of two overlapping platform-reported clicks or views gets credit for the same install.

None of this would matter much if the windows agreed. They don't. So a UA team summing the installs each channel claims in its own dashboard will always produce a number larger than the total the MMP or the app store recorded; that gap isn't fraud, and it isn't a reporting error either. It is what happens mechanically when two platforms both claim credit for a single install inside their own overlapping windows and the MMP has to arbitrate between them.

Why the lookback choice matters more than it looks

A longer lookback inflates a channel's apparent contribution because it captures installs that would have happened anyway, days after the player saw the ad, and hands the ad the credit. A shorter window undercounts channels with a genuinely longer consideration cycle, typically mid-core and strategy titles where a player might see three or four ads across a week before installing. Genre decides which failure you get. A hyper-casual game with an impulsive install cycle suits a short window; a strategy or role-playing title with a longer consideration period looks artificially weak under that same short window.

Platforms set default windows to suit their own auction dynamics rather than a buyer's need for comparability. That's why the same nominal number, a seven-day click window, can mean different things depending on where a platform starts and ends the window and how it treats multiple touches inside it. A studio comparing Meta and TikTok side by side, assuming that matching seven-day windows means matching measurement, is comparing two numbers that only look alike.

A policy a studio can actually enforce

The fix isn't a universal correct window, because none exists across platforms and event types this different. It's a documented policy the studio applies consistently instead of negotiating fresh for every channel comparison, and three elements separate a usable one from an aspirational one.

Pick one lookback standard for cross-channel comparison, usually the shortest common window across the platforms in the mix, and apply it uniformly whenever the weekly review compares channel performance, even while individual platforms keep longer windows for their own optimisation.

Separate the channel dashboard number, which will always run hot because every platform has an incentive to claim broad credit, from the MMP total, which the MMP has deduplicated and which sits closer to the true count of installs. Present both, labelled. Don't let a single blended number stand in for reality; this site drew the same distinction in Attribution Windows for Hybrid-Monetisation Games, where cohort mixing made the mismatch harder to spot.

Revisit the policy whenever a platform changes its default window or its view-through treatment. These change without much notice and quietly shift the size of the mismatch, so a policy written once and never checked against current platform defaults drifts out of date within a couple of quarters.

What the policy does not fix

None of this closes the gap between claimed and actual installs. It makes the gap legible, which is the most a lookback policy can honestly promise in a market where every platform has its own reason to claim more credit than it delivered. Treat the policy as a settled answer rather than a standing discipline and the gap will have quietly widened again by the time a platform revises its defaults, usually without an announcement large enough for a buying team to notice on its own.

Teams that manage this well schedule a short quarterly check against current platform documentation instead of relying on institutional memory of what the windows were when someone first wrote the policy. Platforms change lookback windows quietly and rarely explain why. A policy built on a year-old assumption about them is already drifting.

Related archive reading

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

• Data clean rooms for game marketers
• Android attribution reporting after Sandbox