SKAN 4 and AdAttributionKit, a year in: what buyers actually get back

A year into working with SKAdNetwork 4 and its successor framework AdAttributionKit as the backbone of iOS measurement, it is worth being precise about what these frameworks actually return, because the gap between what a dashboard implies and what the postback contains is still the most common source of bad decisions in a weekly UA review.

AdAttributionKit postback windows are not a single number

SKAN 4 sends up to three postbacks per install, on a tiered schedule rather than a single fixed delay, and each one carries a different level of detail. The first window arrives fastest and carries the most granular conversion value; the later windows arrive slower and carry progressively coarser data. Treat "the SKAN postback" as one event with one number and you are already misreading the framework, because a campaign's apparent performance can shift materially between the first and third window without anything about the campaign actually changing. Any dashboard that collapses all three windows into a single blended metric, without labelling which window contributed which data, hides information a buyer needs.

Coarse values exist because Apple designed for privacy first

Below a certain install volume threshold, SKAN drops from fine-grained conversion values to a coarse value: low, medium, high. That drop is not a bug in an MMP's reporting pipeline. It is the framework working as intended. Privacy thresholds exist to prevent an install cohort from being small enough to be individually identifiable, and the trade-off is that campaigns running below scale, new creative tests, small geos, niche audiences, get materially less signal than a mature top-of-funnel campaign at national scale. Teams that judge a new creative concept's SKAN data against the same confidence bar as an established campaign's are comparing two different instruments, not two results from the same one.

Re-engagement measurement remains the weakest link

Re-engagement attribution under SKAN and AdAttributionKit is meaningfully thinner than new-install attribution: less granular conversion value support, less mature tooling across the MMP layer generally. Treat it as directional, not decision-grade. A team running meaningful re-engagement spend on iOS should be running its own holdout structure alongside the postback data rather than trusting the postback data alone to say whether a re-engagement campaign worked.

What this means for a weekly review

None of this argues against SKAN or AdAttributionKit as a framework, which exist because the alternative was no iOS attribution at all under App Tracking Transparency. It argues for specificity in how the numbers get read. A weekly review should ask which postback window a number came from, whether the campaign sits above or below the coarse-value threshold for its geo and audience size, and whether a holdout backs any re-engagement number in the deck or whether postback data is standing in for one. A dashboard that cannot answer those three questions for a given row is presenting confidence the underlying data does not support.

The framework has not changed materially in the last year, which is itself useful: a stable measurement floor is easier to build a testing discipline around than a shifting one. Building that discipline still falls to the team reading the numbers. The postback's existence does not do it for them.

Related archive reading

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

• Attribution Windows for Hybrid-Monetisation Games
• Designing a SKAN conversion value schema