Campaign-to-install is a chain of claims.

What the lab treats as real

A campaign may claim an install only if three things can be shown: that the person could have seen or clicked the campaign, that a matching identifier survived until first open, and that the attribution window still applied. If any link is missing, the honest label is unattributed — not “organic,” not “brand,” unless those labels are defined in writing.

App analytics teams in Thailand often inherit MMP defaults set during onboarding. Those defaults are not law. They are a configuration. The lab spends most of its time making that configuration visible: click-through lookbacks, view-through lookbacks, re-engagement windows, and whether a reinstall counts as a new install for reporting.

Identifiers without romance

Click IDs are the cleanest story when they persist. Device graphs and probabilistic residue are weaker stories; they can still be used, but they must be named as such in the install graph. SKAN postbacks replace much of the iOS identifier story with a delayed, coarse signal. We do not treat SKAN as a broken MMP. We treat it as a different claim type.

Where teams usually slip

View-through credit on video-heavy networks inflates ownership when the lookback is long and the catalog is cheap. Click flooding makes a campaign look inevitable. Agency screenshots that start after the MMP’s internal deduplication hide collisions. The lab’s rule is simple: if you cannot redraw the path on paper, you may not present the number upstairs.

How this sits next to the intensive

This page is the method. The Install Attribution Intensive is where you practice it on live-shaped exports. If you only need the method written down, start with Field Ledger. If you need a scored reader on your own graph, ask about a residency.