Reading the attribution dashboard
What each tile, panel and column means, how sure each number is, and when a number is withheld instead of shown wrong.
Dashboard → Attribution. The range starts on the last 28 days. Days are counted in your app's time zone. Every range but all time is compared with the same length of time just before it.
Rules for what counts as an install and a match are on How attribution works.
The tiles
| Tile | What it counts |
|---|---|
| Installs | Installs whose install time is in the range. Test devices are left out. |
| Non-organic | Installs with a match, split into Exact, Modelled and Weak. Fraud-flagged installs are left out. |
| Net revenue | Revenue after store commission and tax. |
| Clicks | /l clicks. Bots, link previews and prefetches are not counted. |
| Cost | Spend from connected ad platforms, or cost declared on a link. |
| ROAS (net) | Net revenue ÷ cost. roas_gross is there too, for comparing with an ad platform's own dashboard. |
An install whose link names no media source shows as unattributed_source. It is non-organic, never organic.
Net revenue and net_coverage
Some revenue has no real net figure, for example when a store hasn't reported fees yet. That revenue falls back to gross. net_coverage is the share of gross revenue that has a real net figure. When it is under 100%, the tile says so.
When a number is withheld instead of wrong
Blended ROAS is not shown unless cost data covers at least 98% of installs. Below that you get "partial" and the coverage percentage. CTR is not shown when clicks exceed impressions. Revenue is stored in USD. Spend in more than one currency, or in a currency other than USD, blocks ROAS with a named reason (mixed_spend_currency or currency_mismatch) instead of a quietly wrong total. There is no currency conversion.
Exact-match rate
The share of non-organic installs that are exact, by platform and by link. This is the number to watch: a drop means a link, domain or SDK setting stopped working. iOS has no modelled installs, so on iOS this is exact ÷ (exact + weak).
Conversion rate (CVR)
Installs whose winning touch is a /l click ÷ unique /l clicks in the same range. Bots are left out of both. Installs from other sources, like Apple Ads or organic, are not in the top of the fraction, so CVR is about your links only.
The KPI table
Group by source, campaign, link or day. Expand a source to see its campaigns. Each row has installs split by kind, plus fraud_flagged and reinstalls.
Each row also shows where its cost came from: api (synced from an ad platform), link_declared (CPC, CPM or CPI set on the link, times the clicks, impressions or installs), mixed, or none. Declared cost is used only where no synced row exists for that provider, campaign and date.
assisted counts installs where this source was a contributor, not the winner. It reads null where it can't be worked out. null means "not measured here"; 0 means "measured, and zero".
Charts
Installs per day by kind, net revenue per install, D1 and D7 retention, in-app event volume and assisted installs per source.
- Retention averages are weighted by cohort size, so a day with 3 installs doesn't count as much as a day with 300.
- New counting rules. Charts draw a marker on the day the current attribution rules went live. Numbers before it are frozen and were counted under the old rules.
The assisted-installs panel isn't a range total
It is worked out in the browser from the fetched install sample (200 rows), not from the whole range. Read it as a direction, not a total.
The install explorer and exports
One row per install, with its match type, match kind (exact, weak or modelled), confidence (0 to 1; exact is 1), any fraud rule that flagged it, and its contributors. The same columns are in installs.csv. See CSV exports.