Reinstall or New Install? Why iOS and Android Count Differently
A user deletes your app, installs it again a few weeks later from a friend's invite link, and signs up. Whether that is a new install depends on what is still known about the first time, and iOS and Android leave an app with different things to go on. The same action can add one to Installs on one platform and one to Reinstalls on the other.
Why the label matters
Three numbers depend on telling a first install from a reinstall.
- Install counts. If every returning user counts as new, growth looks better than it is, and so does any rate that divides installs by clicks.
- Attribution. Attribution means crediting an install to the link tap that caused it. A campaign that brings back a lapsed user did not acquire a new one.
- Referral rewards. A referral program pays for new users. If deleting and reinstalling makes someone new again, two people can invite each other for as long as the rewards last.
What is still known after an app is deleted
An SDK works out whether the app was on a phone before from identifiers and stored values, and the two platforms differ on which of those outlast the app.
Android
The device identifier AdTo's SDK reads on Android is ANDROID_ID. Android's reference describes it, on Android 8.0 and higher, as unique to each combination of app-signing key, user and device. The Android 8.0 behaviour notes add that the value does not change on package uninstall or reinstall as long as the signing key is the same. It may change after a factory reset or when the signing key changes.
The package manager also reports when the app was first installed and when it was last updated (firstInstallTime and lastUpdateTime). AdTo's SDK compares the two to tell an update from an install. It does not use them to spot a reinstall: AdTo's documentation lists a reinstall on Android as not detectable on the phone.
A file in the app's own storage is no reliable marker either. Android's Auto Backup restores backed-up app data whenever the app is installed, and always leaves out the cache and no-backup directories. AdTo's SDK keeps its own install record in the no-backup directory, so a restored backup cannot make a fresh install look already registered.
iOS
The matching identifier is identifierForVendor (IDFV). Apple's documentation says it stays the same while the app, or another app from the same vendor, is installed, and changes when the user deletes all of that vendor's apps and then reinstalls one.
So the platform identifiers already disagree. Delete an app and install it again: on Android the id is the same, and on an iPhone with no other app from that vendor it is a new one. A service that relied on the platform id alone would call the first a reinstall and the second a new device.
The other place an iOS app can keep a value is the Keychain. Apple's Keychain services documentation describes it as an encrypted database for small pieces of user data. Neither that page nor Apple's guide to restricting keychain item accessibility says what happens to an app's items when the app is deleted, so a value that outlasts deletion is behaviour an SDK relies on, not a documented promise. What the guide does document is the device boundary: an item whose accessibility setting ends in ThisDeviceOnly is not migrated when another device's backup is restored.
How an attribution service decides
There are two memories to consult: the phone's and the server's. In AdTo, an install is a reinstall when either one says so: the app reported that it was installed on this device before, or the server already has an install with the same device id in the same account.
On an iPhone the SDK can answer from the phone. By default it keeps the first IDFV it reads in the Keychain, as a this-device-only item, and reuses it as the device id. It also keeps a first-install date there. When that date is more than 24 hours older than the app's current Documents directory, the SDK reports the install as a reinstall. AdTo's documentation describes both values as surviving deletion of the app.
On Android only the server can answer. The SDK sends ANDROID_ID as the device id, and the server looks for an earlier install with that id.
The verdicts differ whenever the two memories disagree:
| What happened | iPhone | Android |
|---|---|---|
| The user deletes the app and installs it again, and the server still has the earlier install | Reinstall. The phone says so, and the server knows the saved device id. | Reinstall. The server knows the ANDROID_ID. |
| The same, after you erased that device's data on the server | Still a reinstall. Erasing does not reach the phone. | New install. The only memory was the server's. |
| The same, after the earlier install passed the retention cutoff (400 days by default) and the server no longer holds its device id | Reinstall, for as long as the SDK finds its first-install date. | New install. There is no earlier device id to compare with. |
The ANDROID_ID changed: a factory reset, or a build signed with another key | Does not apply. | New install. The server sees a device id it has never met. |
A debug build and a store build signed with different keys read different ids on the same Android phone, so in testing each looks like a separate device.
AdTo matches a reinstall like any install, and it can be credited to a link. The flag only moves it from Installs, which counts first installs, to Reinstalls. The full order of signals is in How attribution works.
An app update is neither
The day you ship your first build with an attribution SDK, every existing user who updates has a "first launch" as far as the SDK can tell. Counted as installs, they would make your whole audience look new, and each could take credit for a tap that belonged to someone else.
AdTo's SDK labels these on the phone. On Android it is an update when the last update time is more than 24 hours after the first install time. On iOS it is an update when the app's executable was created more than 24 hours after its Library directory. The server never runs its fingerprint match, the scored comparison with recent taps, on an update, although a deterministic signal such as a store referrer still applies. An update counts as neither an install nor a reinstall: the Overview shows app updates on a line of their own, and an update's first identify call, where your app hands over its user id, records no signup lead and no referral. The details are in Install attribution.
What it does to referral credit
AdTo records a referral when a new customer's first identify call runs on an install that is at most 30 days old, is not a reinstall or an app update, and is credited to a referral link. A friend who had the app on that phone before fails the second test, even with a brand-new account and a real tap on the invite.
An iPhone keeps reporting the reinstall for as long as the SDK finds its Keychain values. On Android the answer rests on the server alone: it holds while the server still has the earlier install's device id and the ANDROID_ID has not changed.
Check the flag before the app promises anything:
final install = await adto.getAttributionData();
final referrer = install?.link?.referrerCustomerId;
if (referrer != null && !install!.isReinstall) {
showInvitedBy(referrer);
}
Use that for the screen only. Grant the reward on your server from the referral.converted webhook, a signed request AdTo sends you, because a modified app can show anything. Referral programs has the full flow, and Referral links that credit the right person covers link design.
Reading it in the dashboard
- Overview. Installs counts first installs. Reinstalls has its own card. Conversion divides attributed first installs by clicks. The match confidence panel counts every install, reinstalls and app updates included, so its total is higher than the Installs card.
- Links. A link's Installs number counts first installs, with
+{n} reinstallsunderneath when a new tap on the link won any. - Installs. A row carries a chip for a reinstall or an app update, and the install detail includes reinstall in its verdict. See Installs.
- Customers. A customer's page shows Reinstall as Yes or No. When a link named a referrer and no referral was recorded, Referred by shows the referrer with a not credited chip and the reason. See Customers.
A test phone is the usual surprise: it had the app before, so its referral is refused. On Android, Erase this device in the install detail is enough for the next install to count as new; tap the link again before you reinstall. On the iOS simulator, delete the app and run xcrun simctl keychain booted reset, which resets that simulator's whole Keychain. On a real iPhone, use a phone that never had the app.
What to tell a user or a support agent
| Reason | What happened | A reply you could send |
|---|---|---|
| A reinstall | The phone had the app before. | "Invites reward people who are new to the app. This phone had it before, so the install counted as a return." |
| An app update | The friend already had the app and updated it. | "Your friend was already a user when they tapped the invite." |
| The customer existed before the install | The friend signed in to an account they already had. | "The account is older than this install, so it is not a new signup." |
The other reasons are a signup more than 30 days after the install, a second account on an install that already credited a referral, a link that named the customer themself, and an install whose credit a blocked fraud rule withheld.
Doing this with AdTo
AdTo's Flutter SDK, flutter_adto, labels each first launch as an install, a reinstall or an update, and the server adds its own check by device id. The dashboard keeps the three apart: first installs under Installs, reinstalls beside them on the Overview and on each link, and app updates on their own line.
Reinstalls are still matched to links, so you can see which link brought someone back. Referrals and automatic signup leads are recorded on first installs only, and when a referral is refused the customer's page says why.
Create an AdTo account to see how your own installs split. For the identifiers behind the device id, read Device data and privacy.