Testing & Testers16 min read

Which Play Console Metrics Actually Predict Uninstalls

Play Console reports uninstalls after the fact. Here is which of its metrics move first, how much warning each one gives, and how to read them in order.

Ahmed Akash
Ahmed Akash
Which Play Console Metrics Actually Predict Uninstalls

Need testers who stay active for all 14 days?

TesterBee matches you with 12 real Android testers in under 24 hours. Refund if Google rejects you over tester engagement.

Get 12 testers
On this page
  1. 01The short answer
  2. 02What Google has said about the link to uninstalls
  3. 03Three clocks, not one
  4. 04The leading indicators, in the order they can change a decision
  5. 05Why user loss rate cannot be your early warning
  6. 06Reading the release dashboard in the order Google documents
  7. 07What Play Console will not tell you
  8. 08A watch order for the first week of a release
  9. 09Frequently asked questions
  10. 10What this changes in practice
  11. 11Sources

The short answer

Play Console will tell you that uninstalls rose last week. It will not point at the number that moved the week before. The closest thing to an official answer comes from two Google sources: the 2022 Android Developers post that introduced user-perceived core vitals, and the release dashboard help page that sets out what to check when uninstall events spike.

Read together, they support a simple split. User-perceived crash and ANR rates behave like predictors. User loss rate behaves like a scorecard. One moves within hours of a bad build, the other is defined over a 30-day activity window and cannot move quickly even if it wanted to. Knowing which number you are looking at is most of the skill.

When Play replaced its older stability metrics with user-perceived rates, the post explaining the change made a claim about prediction. The new numbers, it said, are "a better reflection of user experience", and Google added: "we've seen a stronger correlation with uninstalls" (Android Developers Blog).

That sentence deserves a careful read. It does not say crashes cause uninstalls. It says that counting only the crashes a user was actually present for tracks uninstall behaviour more closely than counting every crash your logs record. Cluster totals are therefore the wrong trigger; distinct affected users are the right one.

It also means any figure claiming to convert a crash rate into a churn percentage is not Google's. No such conversion appears in the Android vitals documentation, the release dashboard help page, or the developer quality pages. The relationship is stated as a correlation and left there, which is the honest position: a crash during onboarding and a crash in a settings screen are the same event to a crash counter and completely different events to a person deciding whether to keep the app.

Three clocks, not one

Every figure in the console is averaged over its own window, and those windows sit far apart. A release's crash number can move on the day it ships. Its loss rate is defined using devices that were switched on in the past 30 days, so the same release cannot produce a threshold breach in the same week.

  1. Hour 0 to day 3: per-release user-perceived crash and ANR rates. The Test and release overview shows these against your other releases and switches to hourly granularity for the first few days of a new release, then back to daily (Play Console Help).
  2. Day 1 to day 7: the Android vitals dashboard, updated daily, holding 90 days of history in the console and three years through the Reporting API (Play Console Help).
  3. Until the book closes: the release dashboard keeps reporting a release for seven days after you stop increasing its rollout, one day after a halt, or one day after a newer release starts, whichever comes first (Play Console Help).
  4. Day 30 and after: user loss rate, the only uninstall number Play publishes a threshold for, evaluated against an installed audience whose devices must have been active in the past 30 days.
Four windows in one release: the crash rate moves within hours, the release dashboard closes within days, and the loss rate needs a month before it can confirm what the crash rate already suggested.

The practical rule that falls out of this: a decision that has to be made during a rollout cannot be based on a metric defined over 30 days. If you are deciding tonight whether to widen a rollout, the loss rate has nothing to say yet.

The leading indicators, in the order they can change a decision

Per-release crash and ANR rate

This is the fastest number Play gives you and the one the release dashboard is built around. You pick a primary release and a comparison release from two drop-downs, and every card is drawn from that release's first rollout day onward, with each rollout milestone marked on the timeline.

One trap sits inside the card. Crashes and ANRs are shown per thousand devices so releases of different sizes can be compared, and Google states that when installs are too low for normalisation to mean anything, the line is drawn dotted. Early in a small rollout the dotted line is telling you the number is not evidence yet, not that the build is clean.

A single device model crossing the per-model threshold

Play's thresholds are 1.09% for user-perceived crashes, 0.47% for user-perceived ANRs, and 8% on any single phone model (Android Developers). That last number is where a prediction gets specific, because a per-model spike is usually one OS version, one chipset or one OEM background-process policy rather than a mistake in your own code.

Our reading of campaign data, and not a Google statement, is that per-model breaches also cluster geographically: the same handset dominates in a few markets, so a breach on one model often means a whole country is churning. What Google documents is narrower and firmer: if a title exceeds the per-model threshold, Play may reduce its visibility for users on that model and steer them toward other apps.

Slow start and low-memory kills

Two of the most useful early signals carry no threshold at all. Start-up time, reported as time to initial display, is an alerting metric in Android vitals but not a core vital, so a startup regression will never trip Play's visibility rules. It only shows as a trend, which means nobody gets an email about it.

Low memory kills work the same way. The dashboard reports a user-perceived LMK rate, meaning kills that happened while your app was on screen or running a foreground service, and again Google publishes no bad behaviour threshold. A user-perceived LMK looks exactly like a crash to the person holding the phone: the app disappears. It is counted separately, so a product with a clean crash rate can still be losing users to memory pressure.

These two metrics are the answer to a specific complaint pattern: reviews say the app is slow or closes itself, and your stability numbers look fine. If you are triaging that pattern, our guide to what each Android vitals number means covers where each metric lives and which threshold it answers to.

Excessive partial wake locks

This one is a core vital with a published threshold, and it is stated as a share of battery sessions: partial wake locks running longer than three hours in more than 5% of them, currently in beta (Android Developers). It affects discoverability, and it is a genuine early warning for one specific uninstall reason, battery drain, that no crash report will ever capture. What Google does not publish is any correlation figure tying wake locks to uninstalls, so treat it as a complaint signal rather than a predictor with a known strength.

Why user loss rate cannot be your early warning

Two bars sit on Play's core value page (Android Developers): user loss rate below 5%, and daily active users divided by monthly active users above 8%. The same page also requires enough installs and engagement for the assessment to cover at least 24 days out of 30. Those figures decide eligibility for certain Play surfaces and tools, which is why they turn up in planning documents.

They are the wrong numbers to page someone about. A loss rate is a ratio whose denominator is the installed audience, and Google's Statistics page defines that audience using devices that have been switched on in the past 30 days. A release-day uninstall spike therefore enters a numerator that is being divided by a base still dominated by a month of normal behaviour. The signal is real and it is diluted by construction.

There is a second, quieter problem: the console holds 90 days of vitals history. Compare this month's loss rate to the same month last year and the console has nothing to show you, because that data has aged out. If that comparison matters for your reporting, export it monthly. Our guide to reducing uninstalls and improving 30-day retention covers the definitional traps in that ratio, including the difference between a user who uninstalled from every device and one whose phone simply stopped being used.

Reading the release dashboard in the order Google documents

The release dashboard help page contains the nearest thing to an official triage sequence, and it is short. If you notice a spike in uninstall events after an update, Google says to review crash and ANR data for an above-average trend, then review the average rating for the release along with its one and two star reviews. It adds a second rule: if crashes per thousand devices have risen significantly against a successful previous release, consider pausing the release to investigate.

Two mechanics make that sequence workable. First, uninstall events on the dashboard are split into uninstalls after a new install and uninstalls after an update. That split is the most informative line on the page. After-update uninstalls are attributable to the release you just shipped; after-new-install uninstalls are an onboarding failure that would have happened with any build. The fixes are not interchangeable.

Second, reviews are filtered to one and two star reviews for the latest release by default. That is the only place in Play Console where a user tells you, in words, why they left, and it is the reason Google's triage order puts ratings after stability rather than before it: the reviews explain a spike the stability numbers have already quantified.

Then the window closes. A halted release keeps reporting for a day; a release whose rollout you stopped increasing keeps reporting for seven; and a newer release ends the previous one's data immediately. If you need the numbers for a post-mortem, screenshot or export them before the next release goes out, because the dashboard is a live comparison tool, not an archive.

Halting also became more useful than it used to be. A fully rolled-out release can now be halted on any track except internal testing, from the web console, the mobile app or the Publishing API, and the previous version automatically takes its place for new and eligible users. Two limits are worth knowing in advance: you cannot halt the first release on a track, and a previous release that carries a policy violation cannot be used in its place (Play Console Help).

What Play Console will not tell you

The gaps matter as much as the metrics, because each one is a place where a confident dashboard turns into a guess.

  • No reason per uninstall. Play reports counts, not causes. The closest thing to a stated reason is the one and two star review text for the release, which is why it belongs in the triage order.
  • No per-user attribution. Nothing connects a specific crash report to a specific uninstall. The dashboard can tell you that a release lost users and crashed more; it cannot tell you the same person did both.
  • No user loss rate through the API. The Reporting API's documented metric sets cover ANR rate, crash rate, excessive wake-up rate, slow start rate, slow rendering rate, stuck background wake lock rate, low memory kill rate, memory usage, error counts and anomalies (API reference). There is no metric set for user loss or DAU and MAU, so the two core value numbers stay a manual console read, or a monthly export.
  • No email about a loss spike. Notifications are only available for anomalies, and anomaly detection is supported on the ANR, crash, excessive wake-up and stuck wake lock metric sets. A quiet month of rising uninstalls generates nothing.
  • No coverage of installs outside Play. Vitals exclude technical issues on uncertified device models and on versions of your app that were not installed through Google Play, so a sideloaded build's crashes never enter the comparison.

A watch order for the first week of a release

This is the sequence that follows the clocks above rather than fighting them. Each step uses a number that is fast enough to act on at that moment.

  1. Before rollout: open the release dashboard and set the comparison release to the last build you were happy with. The default comparison is all releases, which hides a regression inside your own average.
  2. Hour 0 to day 3: read the per-release user-perceived crash and ANR rates in the Test and release overview, at hourly granularity, and ignore any dotted normalised line until the install count supports it.
  3. Day 1 onward: break the vitals down by device model and check each model against the per-model bar, not just the overall rate against the overall bar.
  4. If stability is clean but complaints are not: check start-up time and the user-perceived low memory kill rate. Neither carries a threshold, and either explains an app that feels broken without failing.
  5. When uninstall events rise: separate after-update from after-new-install before changing anything. The first points at the build, the second at onboarding.
  6. Same day: read the one and two star reviews for the release, then capture every number you will need later, because the dashboard stops reporting this release within days of the next one shipping.

Frequently asked questions

Does a high crash rate always lead to uninstalls?

No, and Google does not claim it does. The company's stated position is that user-perceived rates correlate more strongly with uninstalls than the older metrics did. Correlation varies by app and by where the crash happens: a crash at first launch costs far more than one behind a settings screen that most users never open.

Which Play Console metric changes first after a bad update?

The per-release user-perceived crash and ANR rates. The Test and release overview reports them hourly for the first few days of a new release before switching to daily, which is why it is the only screen fast enough to justify halting a rollout on the same day.

Can I see uninstalls for one specific release?

Yes, on the release dashboard, and the number is split into uninstalls after a new install and uninstalls after an update. Only the second is attributable to the build you shipped. The dashboard keeps that release's data for seven days after you stop increasing its rollout.

My loss rate has not moved after a crash spike. Is the data broken?

Probably not. Loss rate is defined against an installed audience of users whose devices were active in the past 30 days, so a single bad week is diluted by a month of history. The crash rate is the number that records the event; the loss rate is the number that records the consequence, later.

Is my DAU/MAU ratio good?

Google does not publish a benchmark for it, so there is no "good" figure to compare against. What exists is an eligibility bar on the core value page (Android Developers): daily active users divided by monthly active users above 8%, alongside a user loss rate below 5%. Treat those as minimums for Play surface eligibility, not as targets to celebrate.

Does Play Console email me when uninstalls spike?

No. Console notifications cover anomalies only, and anomaly detection exists for the ANR, crash, excessive wake-up and stuck wake lock metric sets. Nothing in the console will notify you about user loss, so a monthly look at the loss rate is a process you have to own.

Can I pull user loss rate from the Play Developer Reporting API?

Not from a documented metric set. The API exposes vitals metric sets and error data, and the core value metrics are not among them. The practical workaround is a scheduled export from the Statistics page, or the monthly reports Play already generates.

Are crashes the main reason people uninstall?

Play does not publish a ranking, so anyone who gives you one is estimating. What the console can show is a sequence: a release crashes more than its predecessor, then uninstall events after update climb. That sequence is strong enough to act on. The reverse claim, that crashes are the largest cause of loss overall, is not supported by published data.

What this changes in practice

Most teams watch the loss rate because it is the only uninstall number with a published threshold, then discover it tells them about a problem a month after the release that caused it. Reading the console on its own clocks removes that delay: crash and ANR per release for the decision, per-model rates for the location, start-up and memory kills for the shape of the complaint, and the split between after-update and after-new-install uninstalls for the cause.

One habit is worth adding before the next release: test the build the way the users who leave will experience it, on the devices that cross the per-model line. A pre-release pass on real hardware in the markets you ship to, which is what the pre-launch report is for, converts a per-model surprise into a fix you can ship in the same sprint.

Sources

Ahmed Akash
Ahmed Akash·Senior Full Stack Developer & Founder of TesterBee

Senior full stack developer specializing in Laravel, React, and Firebase. Helped 1,300+ Android developers pass Google Play Closed Testing and publish apps. Expert in Play Console, production access, and Android app publishing.

November 21, 2026 · 16 min readLinkedInGitHub

Get 12 testers into your closed test in 24 hours

Matched Android testers who open your app every day of the 14-day window, so your production access review has real engagement to look at.

Start your closed test