The short answerCrash-free users and crash-free sessions describe different stability experiences. One asks how many tracked users avoided a crash; the other asks how many sessions avoided one. Use the same window and reporting definitions, and inspect affected versions before making growth decisions.

Understand the provider's units

Firebase Crashlytics defines a user for these metrics as an app installation on a device, not necessarily a unique person across devices. Its crash-free metrics concern fatal events under the documented conditions. Check collection coverage and supported SDK behavior in the current guidance.

Avoid averaging daily percentages when the provider defines an aggregate over the chosen window. The same installation can appear on several days, so adding daily user counts can duplicate it.

Worked example: the metrics tell different stories

Illustrative example: 100 tracked installations generate 700 sessions in a window. Ten installations experience a crash, and twenty sessions end in a crash. Crash-free users is 90%. Crash-free sessions is approximately 97.14%.

MetricIllustrative calculationResult
Crash-free users1 − 10 ÷ 10090%
Crash-free sessions1 − 20 ÷ 700About 97.14%

This is original example data, not a real app result. The stronger session percentage does not erase the experience of the ten affected installations.

Inspect the version and task

Separate a newly released version from older versions where the data supports it. Identify whether crashes occur during onboarding, payment or another critical task. An aggregate can conceal a severe problem affecting a smaller segment.

Use comparable windows and note collection limits. More sessions give a user more opportunities to experience a crash, so a longer user-level window can change the meaning of the percentage.

Connect stability to acquisition decisions

If new customers encounter a broken first task, buying more installs can amplify the problem. Use stability as a relevant safeguard while investigating the supported workflow. Do not claim that repairing a crash necessarily creates a particular retention lift.

Review affected cohorts and actual first-use outcomes with the onboarding guide, then decide whether to repair, restrict exposure or resume the planned growth test.

Primary-source context: Firebase Crashlytics: crash-free metrics. Checked 2026-10-13.

Ayoub Mouhachtt
Growth & performance marketing. Explore my working background.

An AI-assisted educational guide commissioned for Ayoub’s Growth Notebook. Worked examples are illustrative; they are separate from the portfolio’s client evidence. How these guides are prepared.