Android vitals is the Google initiative and Play Console dashboard that measures the technical quality of an Android app on real user devices, covering crashes, freezes, startup speed, rendering smoothness and battery use. In mobile performance work, it is the authoritative external view of how an app behaves in the field, because the data comes from the operating system rather than from an SDK the developer installs.
The program matters beyond diagnostics: Google Play uses core vitals thresholds to judge whether an app shows bad behavior, and apps that exceed them can receive reduced store visibility and warning labels on their listing. Android vitals therefore sits at the intersection of engineering, product management and distribution strategy within the wider Mobile discipline.
How Android vitals collects and aggregates data
Android vitals collects data passively from the Android platform on devices whose owners have opted in to share usage and diagnostics data. No library has to be added to the app, which means the metrics cover every release, including builds where an in-house crash reporter failed to initialize. The platform records events such as process crashes, Application Not Responding dialogs, wake lock acquisitions and frame timings, then uploads anonymized aggregates to Google.
Play Console presents the results over a rolling 28-day window, broken down by app version, Android version, device model, country and other dimensions. Most rates are user-perceived and normalized per daily active user, so a single device that crashes a thousand times counts once for that day. This design rewards fixing problems that touch many people rather than chasing noisy outliers.
The table below summarizes the stages of the data pipeline and what each one contributes.
| Stage | What happens | Practical consequence |
|---|---|---|
| On-device capture | The OS logs crashes, ANRs, frames and wake locks | No SDK or code change required |
| Opt-in upload | Aggregates sent from consenting devices | Sample is large but not complete |
| Aggregation | Events grouped per user and per day | Rates reflect affected users, not raw counts |
| Benchmarking | Values compared with thresholds and peer apps | Bad behavior flags and peer percentiles |
| Reporting | Play Console views and the Reporting API | Dashboards, alerts and automated exports |
Core vitals and other Android vitals metrics
Google separates metrics into core vitals, which directly affect Play store treatment, and other vitals, which serve as diagnostic signals. The two core vitals are the user-perceived crash rate and the user-perceived ANR rate. Both are evaluated overall and per device model, because an app can look healthy globally while failing badly on one popular phone.
The following table lists the principal metrics, with thresholds stated approximately because Google revises them over time and publishes the current values in Play Console.
| Metric | What it measures | Bad behavior threshold (approximate) |
|---|---|---|
| User-perceived crash rate | Share of daily users who saw a crash while the app was visible | About 1.09% overall, 8% per device |
| User-perceived ANR rate | Share of daily users who saw an ANR while engaging with the app | About 0.47% overall, 8% per device |
| Excessive partial wake locks | Sessions holding wake locks for long cumulative periods in the background | Defined by Google, revised periodically |
| Slow startup | Cold, warm and hot starts above time limits | Roughly 5 s cold, 2 s warm, 1.5 s hot |
| Slow rendering and frozen frames | Sessions with many slow frames or frames over 700 ms | Diagnostic, no store penalty |
The distinction between user-perceived and total rates is significant. A crash in a background service that the user never notices still appears in the total crash rate but not in the user-perceived rate, which is the figure that drives store consequences.
Android vitals compared with in-app monitoring tools
Android vitals complements, but does not replace, crash analytics SDKs and custom telemetry. Platform data is unbiased and comprehensive in coverage, yet it offers limited context: stack traces are available, but custom logs, breadcrumbs and user journeys are not. Mature teams therefore use Android vitals as the source of truth for store standing and an SDK for root-cause analysis.
| Aspect | Android vitals | Crash analytics SDK |
|---|---|---|
| Installation | None, built into the platform | Library added to the app |
| Coverage | All opted-in devices, every build | Only builds with a working SDK |
| Context | Stack traces and device dimensions | Logs, custom keys and user flows |
| Latency | Typically hours to a day | Near real time |
| Store impact | Defines bad behavior status | No direct store effect |
The metrics also connect to neighboring concepts in the Topic. App startup time is measured directly, jank surfaces through slow rendering, and misuse of wake locks or ignorance of Doze mode and background execution limits shows up as battery-related vitals.
Using Android vitals in a release process
Teams obtain the most value from Android vitals when they tie it to staged rollouts. During a percentage rollout, comparing the new version with the previous one by device model and Android version reveals regressions before the release reaches everyone. The Play Developer Reporting API allows the same data to feed internal dashboards and automated release gates.
The decision table below shows common situations and the recommended response.
| Situation | Recommended action | Reason |
|---|---|---|
| Crash rate rises after release | Halt the staged rollout and inspect top clusters | Limits the number of affected users |
| ANR spike on one device model | Profile main-thread work on that hardware | Per-device thresholds can trigger penalties |
| Slow cold start trend | Defer initialization and use baseline profiles | Startup affects retention and ratings |
| Wake lock warnings | Move work to WorkManager with constraints | Respects Doze and background limits |
| Peer percentile falling | Prioritize the weakest vital in planning | Relative quality shapes user expectations |
Teams should also set internal targets below Google thresholds, for example a crash rate well under 1%, because approaching a limit leaves no margin for a bad release. Monitoring per device model is essential when the user base is concentrated on a few popular phones.
Limitations and common misreadings of Android vitals
Android vitals has structural blind spots and delays that practitioners must account for. The data covers only users who opted in, it arrives with a delay, and low-traffic versions or device models may show too few users for reliable rates. Interpreting it without these caveats leads to poor decisions.
- Small samples mislead: a rate computed from a few hundred users can swing sharply from day to day.
- Native crashes need symbols: without uploaded debug symbols, native stack traces are hard to read.
- Fixed crashes linger: the 28-day window keeps old problems visible after a fix ships.
- ANR causes are indirect: the reported trace shows where the main thread waited, not always why.
- Diagnostic vitals still matter: slow rendering carries no penalty but affects ratings and retention.
A further misreading is treating a green dashboard as proof of quality. Android vitals does not measure functional bugs, network latency or business logic errors, so it should be one input among several in a mobile performance strategy.
Explained without jargon
Android vitals is like a health report card for an app, written by the phone itself. Every time an app crashes, freezes or drains the battery on a phone whose owner agreed to share diagnostics, Android notes it quietly. Google collects these notes from many phones and shows the app maker a summary.
The key point is that Google also reads this report card. If too many people see crashes or frozen screens, the app may be shown less often in the store. Developers watch these numbers so they can fix problems quickly and keep their app easy to find.
A typical use case
A team behind a regional banking app ships version 6.2 through a staged rollout at 10% of users. Two days later, Android vitals shows the user-perceived ANR rate climbing from 0.2% to 0.9% overall, and to more than 10% on one mid-range phone model with 3 GB of memory. The overall figure already exceeds the bad behavior threshold.
The engineers pause the rollout and inspect the ANR clusters. The traces point to a new fraud-check module that reads a large encrypted file on the main thread during login. They move the work to a background coroutine and add a short loading state.
Version 6.2.1 resumes the rollout, and within a week the ANR rate returns to 0.2%. Because the problem reached only a tenth of users and was fixed quickly, the 28-day average never triggered store penalties, and the team adds an automated ANR gate to future releases.