3 Mar 2026 · Privacy windows

Reading SKAN postbacks without guesswork

SKAN is not a broken MMP. It is a different kind of claim. The postback arrives late, the conversion value is coarse, and many genuine installs never produce a row you can screenshot. Teams that treat every missing postback as a failed campaign are guessing. Guessing is not app analytics.

In Thailand, iOS often sits beside a much larger Android catalog. That imbalance tempts growth leads to ignore SKAN until a brand campaign needs an App Store story. Then the postbacks look “too low,” and someone asks the intensive whether Apple is hiding their users. Usually Apple is doing what the privacy window advertised.

What a postback can support

A postback can support that an install-like conversion occurred inside a campaign’s eligible window, at a coarse value you configured. It cannot support a user-level journey. It cannot support a precise CPI comparable to an Android MMP row without a translation step you write down. If you skip the translation step, you are mixing claim types in one table — deterministic click-ID installs next to privacy-window counts — and hoping finance will not notice the units changed.

Conversion values are where guesswork usually enters. Teams map every in-app event onto the 0–63 scale until the mapping is a novella. Then they are surprised when noise and lock windows make the story unreadable. A smaller mapping that you can explain in one paragraph will survive a controller’s questions. A maximal mapping will not.

Missing is allowed

Crowd anonymity thresholds, delayed postbacks, and campaign eligibility rules all create absences. Absence is not evidence that the campaign did nothing. It is evidence that SKAN will not narrate that install. The honest label in an install graph is “SKAN-silent,” not “organic,” unless you have a separate, named rule for organic.

Vault Nodecore’s SKAN studio block spends most of its time on what you may not say. If your catalog is Android-heavy, you may skim this note and still need the click-ID and fraud weeks more. If you do run meaningful iOS spend, print your conversion-value map, your lock window, and one week of postbacks side by side before you argue about CPI. That is reading without guesswork. The lab page puts SKAN next to click IDs as a distinct claim type, which is how it should stay in your reports.