Crash reporting that earns its keep: Crashlytics, ANRs and a written rollout bar
Users uninstall instead of reporting crashes. Grouped issues, crash-free metrics, custom keys and new-issue alerts close the gap - if symbols upload automatically and rollouts gate on a number.
Users do not report crashes; they uninstall. By the time a one-star review mentions "keeps closing", the bug has been costing you users for days. A crash reporter closes that gap: it catches every fatal exception and app-not-responding event in the field, groups them into issues, and tells you which releases are stable enough to promote. Firebase Crashlytics is the most common choice on Android and iOS, and its concepts generalise to whatever tool you pick.
What the reporter actually gives you
Crashlytics is a real-time crash reporter. In practice that means four things matter day to day:
- Crash-free metrics. The headline number is the share of users and sessions without a crash, per release. It is the number to watch in the hours after each rollout, and the honest answer to "is this version safe to promote to 100%?"
- Grouped issues with stack traces. Individual crash events are deduplicated into issues ranked by affected users, so you fix the crash hurting 2,000 users before the one hurting three.
- ANR debugging on Android. App-not-responding events - the app frozen long enough for the system to offer to kill it - are surfaced with their own traces. ANRs often outnumber hard crashes on Android and drive just as many uninstalls.
- Context you attach yourself. Custom keys (user tier, experiment bucket, screen name) and custom logs travel with each report, turning "NullPointerException in CheckoutFragment" into "only on the new checkout, only for users in the B variant". Non-fatal exceptions can also be recorded explicitly, which is how you track handled-but-wrong states like failed payments.
Setting it up so the data is trustworthy
Upload your symbol files. Release builds strip or obfuscate symbols (ProGuard/R8 mappings on Android, dSYMs on iOS, debug symbols for native code). Without the mapping upload step, stack traces are memory addresses. Automate the upload in the same CI step that builds the release - a trace you cannot read is a bug report with the headline removed.
Gate rollouts on crash-free rate, not vibes. Both stores ship staged rollouts. Start each release at a small percentage and promote only when the crash-free user rate holds. Define the bar in writing before you need it - for example "promote past 10% only if crash-free users are at or above 99.5% and no new top issue appeared" - so the decision at 11 pm is mechanical.
Alert where the team lives. Crashlytics supports alerts by email and in-console, with basic integrations for Slack, Jira and PagerDuty, plus custom notification channels. A regression found by an alert at 9 am costs a fraction of the same regression found by reviews a week later. Route new-issue alerts to the channel someone actually reads.
The privacy line
Crash reports contain device state and anything you attach in custom keys. Keep user-identifying data out of keys and logs unless you have a documented reason and disclose it in your privacy policy. If you use an identifier to correlate crashes with support tickets, make it an opaque ID, not an email address. Both store policies and privacy law treat crash data as personal data when it can be tied to a person.
What good looks like
A healthy setup is boring: symbols uploaded automatically with every build, a written crash-free bar per rollout stage, new-issue alerts in the team's chat, and a weekly ten-minute triage of the top issues. When a release regresses, you know within hours, you halt the staged rollout from the console, and the fix ships with the exact trace attached. That loop - not any single feature - is what a crash reporter buys you.
Sources
Prepared with AI assistance and reviewed before publication. Features described from Google's Crashlytics documentation, checked on 19 November 2026. The example 99.5% rollout bar is an illustration, not a Firebase recommendation - set your own threshold.