▸Example AppsFlyer call (the "before")
import AppsFlyerLib
// ATT prompt path — required before AppsFlyer reads IDFA:
ATTrackingManager.requestTrackingAuthorization { _ in
AppsFlyerLib.shared().start()
AppsFlyerLib.shared().logEvent("af_purchase", withValues: [
AFEventParamRevenue: price,
AFEventParamCurrency: "USD",
AFEventParamContentId: sku,
])
}
Most analytics SDKs back the unsent event queue with SQLite or UserDefaults / SharedPreferences — so a phone that's been confiscated, jailbroken, or restored from backup still contains analytics state. Respectlytics's queue is RAM-only, flushed on a 30-second timer; unsent events on force-quit are lost by design, in exchange for zero on-device forensic surface.
☑Remove AppsFlyer cleanly
-
1
Remove the AppsFlyer SDK from your build (
AppsFlyerFramework/af-android-sdk/react-native-appsflyer/appsflyer_sdk) -
2
Remove
AppsFlyerLib.shared().start()andlogEvent(...)call sites -
3
Re-check your
Info.plistforNSUserTrackingUsageDescription— if no other SDK needs ATT, remove it (Apple flags apps that ship the key without code that callsATTrackingManager) -
4
Re-check your Android merged manifest for
com.google.android.gms.permission.AD_IDand remove the corresponding<uses-permission>if no other SDK contributes it -
5
Plan how you'll attribute installs without AppsFlyer — Apple SKAdNetwork + Google Play Install Referrer (both first-party, no SDK needed) cover most cases
⇋AppsFlyer vs Respectlytics — ram-only event queue
| AppsFlyer | Respectlytics | |
|---|---|---|
| Event queue persistence | SQLite / UserDefaults / SharedPreferences | In-memory ring buffer |
| Disk usage for analytics | 0.5–10 MB typical | 0 bytes |
| Forensic data on jailbroken / rooted devices | Persistent identifiers + queued events | None |
| Survives force-quit before flush | Yes | No (events lost — by design) |
❓Frequently asked questions
Doesn't this reduce data quality?
Marginally — typical force-quit-before-flush event loss is 0.5–2% depending on platform. For aggregate metrics (funnel rates, feature adoption, release deltas) this is invisible. For per-event reconciliation it would be a problem, but per-event reconciliation isn't a use case Respectlytics supports.
What's the actual flush cadence?
30 seconds by default, plus a flush on applicationDidEnterBackground (iOS) / onPause (Android). Most events reach the network within seconds of being fired.
Is this safe for crash analytics?
Crash analytics is a separate concern — use Sentry, Crashlytics, or Bugsnag (with their own crash-aware queues). Respectlytics is product analytics; crash data has different recoverability requirements and lives in different tools.
Why is this a privacy feature?
Devices that are jailbroken, rooted, restored from backup, or forensically imaged routinely surface analytics artifacts — distinct_ids, queued events, user properties — that survive uninstall in some cases. RAM-only storage moves the dump-recovery surface to zero.