App analytics — where the numbers come from
An app with a website is counted in at least four places that never talk to each other: App Store Connect, Google Play, a web analytics tool and the server or CDN. Each counts something slightly different, at a different time of day, with its own blind spots. This is the map we drew while wiring up our own apps — every factual line links to the official page it was checked against on 16 November 2026.
Keys stay home
🍎 App Store
Two separate datasets: sales reports count transactions, analytics reports count what people did.
Sales reports — the complete count
- One row per app, product type, country and day; the Units column is the count.
- The product type separates first downloads (1, 1F, 1T; F1 on the Mac) from re-downloads (3, 3F) and updates (7, 7F, 7T; F7) — keep only first downloads in “new users”.
- A day runs midnight to midnight Pacific time; the daily report generally appears the next day, by about 8 a.m. Pacific.
- The API asks for your vendor number (shown in Payments and Financial Reports).
Analytics Reports API — what people did
- You create a report request per app. ONGOING keeps producing daily reports; ONE_TIME_SNAPSHOT goes back as far as data exists, once (one snapshot request a month).
- Categories: APP_STORE_ENGAGEMENT (impressions, product page views), COMMERCE (downloads), APP_USAGE (installations, sessions, active devices, deletions), PERFORMANCE (crashes), FRAMEWORK_USAGE.
- Report files are deleted after 35 days — download and keep them if you want long history.
- Usage metrics come only from people who opted in to share data with developers, so they undercount; below 5 users they are not generated at all. Detailed reports add a little random noise and drop values from fewer than 5 users.
- Only an API key with the Admin role can create report requests — the key’s role counts, not yours.
Campaign links, in-app events, custom product pages
- Generate a campaign link in App Analytics → Acquisition → Campaigns: it adds your account’s provider token (pt=) and your campaign token (ct=, up to 30 characters). A campaign shows up after 24 hours and at least 5 first-time downloads.
- Engagement data is split by page type, so a custom product page or an in-app event page gets its own views.
- Downloads carry a source: search, browse, app referrer, web referrer, campaign.
🤖 Google Play
No single analytics API: installs live in files, quality in one API, reviews in another.
Bulk reports in Cloud Storage — installs and uninstalls
- Every developer account has a Cloud Storage bucket (its name starts with pubsite_prod_rev) holding monthly CSV files per app and dimension — overview, country, device and more. They are UTF-16 encoded.
- Data is captured daily but posted a few days later, so the latest days are always missing.
- Daily user installs count Google accounts; daily device installs count devices; “installs on active devices” is how many devices have the app now.
- Store performance files hold listing visitors and acquisitions — their ratio is the listing conversion.
- The bucket address is in Play Console → Download reports → Statistics → Copy Cloud Storage URI.
Developer Reporting API — crashes and ANRs
- Daily crash rate and ANR (“app not responding”) rate, plus the user-perceived versions that Google uses for store visibility.
- The data comes from users who share usage and diagnostics. With too few of them Play simply says there are not enough data points — empty means “too few users”, not “no crashes”. Google publishes no exact threshold.
Publishing API — reviews, and a trap
- Written reviews can be read without opening an edit.
- Track details need an edit, and Google’s rule is blunt: creating a new edit invalidates any edit already open, and committing one invalidates all others for the app — for example a build upload running at the same time. Do not poll edits from a dashboard.
- No Google API tells you that a release is “in review” or “approved”.
Campaign tags on Play links
- A Play link takes a referrer parameter with URL-encoded UTM tags (512 characters at most), e.g. …&referrer=utm_source%3Dsite%26utm_campaign%3Dhero.
- The app can read them once after install through the Play Install Referrer API.
- In Play Console, UTM source and campaign appear as filters in Grow users → Store performance → Conversion analysis, for the “ads and referrals” traffic source.
🌐 The website
Two views of the same visits: one in the browser, one on the server. Real people sit between them.
Google Analytics 4 — people and clicks
- A cookie holds a random client_id per browser: a “user” is a browser, not a person.
- Device category gives the mobile share.
- Enhanced measurement records outbound clicks (a click event with outbound = true — taps on store badges) and file downloads without extra code.
- Its file-download list includes .zip and .exe but not .apk or .dmg: count app files from the server side or send your own event.
- Ad blockers and declined consent hide visits, so GA4 undercounts. Event data retention defaults to 2 months; raise it to 14.
Cloudflare — every request at the edge
- HTTP analytics counts requests on the server side: every page and file, with or without JavaScript — bots and retries included, so it overcounts people.
- Web Analytics is a separate beacon in the browser that uses no cookies or local storage; it keeps six months of data.
- At high volume the data is sampled, and HTTP analytics keeps about a month on every plan — so take a daily snapshot if you want long trends.
🔗 Joining it up: site → store → app
No identifier survives the whole trip. The analytics cookie stays in the browser, and Apple and Google install the app inside their own stores without telling you who tapped your link. Since iOS 14.5 the advertising identifier is all zeros unless the user accepts the App Tracking Transparency prompt, and in the EU personal tracking needs consent.
By channel — no personal data
- Tag every link with the page or button it sits on: ct= on App Store links, UTM in the Play referrer, ?src= on direct file downloads.
- Each source reports counts per tag per day; join the aggregates on channel and date.
- The result is a funnel per channel — visits → clicks → installs — with no cookie banner and no policy change.
Per user — what it would take
- A site ID carried into the app: a deep link on iOS, the Install Referrer on Android.
- The app sends an install ID plus that site ID to your own backend, which joins them.
- A consent banner, updated privacy policies, App Store privacy labels and Play Data safety forms — and a release of every app.
- At small volumes channel-level numbers answer the same questions, so start there.
⚠️ Why two honest numbers disagree
Time zones
Apple days and Play vitals follow Pacific time, other sources use UTC or your own zone. One evening install can land on different dates.
Lag
Yesterday is usually incomplete and recent days can be revised. Compare complete weeks.
Download ≠ install ≠ user
A download may never be opened; Play counts accounts and devices separately; only opted-in data says “active”.
Re-downloads and updates
Neither is a new user. Keep both out of acquisition.
Partial deletions
iOS deletions come only from opted-in users; Play uninstalls are complete.
Conversion denominators
Downloads ÷ impressions and downloads ÷ page views are different rates — say which one.
Small numbers
At three installs a day one tester doubles the chart, and privacy thresholds hide small rows. Read weeks.
Bots on file downloads
Server logs count crawlers fetching your APK; GA4 does not see .apk or .dmg at all without a custom event. Use the server count, filtered.
Double counting
One person on iPhone and Android who also takes the APK is three installs. “New users” across stores means installs, not unique people.
Your own traffic
Testing from your own phone inflates small numbers. GA4 can exclude internal traffic; stores cannot.
🛠 Try it yourself
- App Store Connect → Users and Access → Integrations → App Store Connect API: read the Access column of each key. That role, not yours, decides what the key may do.
- App Store Connect → your app → Analytics: take impressions, product page views and downloads for 30 days and compute both conversion rates by hand.
- Play Console → Download reports → Statistics: download this month’s installs CSV (it is UTF-16) and explain why daily user installs and daily device installs differ.
- GA4 → Admin → Data streams → your web stream → Enhanced measurement: check that outbound clicks and file downloads are on, then tap a download on your phone and find it in Realtime.
- Build one App Store campaign link and one Play link with UTM tags for the same button; a few days later find that campaign in both consoles.
📖 Glossary
| Term | Meaning | Where it lives |
|---|---|---|
| Impression | The app was shown in a list, search result or referral | App Store analytics |
| Product page view | Someone opened the store page | App Store analytics |
| First-time download | An account downloads the app for the first time | App Store sales and analytics |
| Session | The app in the foreground for a stretch of time | App Store analytics (opt-in) |
| User installs | Google accounts that installed, counted once | Play bulk reports |
| Device installs | Devices that installed | Play bulk reports |
| Installs on active devices | Devices with the app currently installed | Play bulk reports |
| Listing acquisition | An install by someone who visited the listing | Play store performance |
| Crash / ANR rate | Share of daily users who hit a crash or a frozen app | Play Reporting API |
| client_id | The random per-browser ID in the GA4 cookie | GA4 |
| pt / ct | Provider token / campaign token in an App Store link | App Store links |
| referrer (UTM) | Campaign tags carried through a Play link | Play links, Install Referrer |