Mobile App Analytics: What Events and Metrics Should You Track?
A founder's framework for app metrics after launch: acquisition, activation, retention, revenue and quality, with each tool's definitions and why they disagree.
Which metrics should founders track after an app launches?
Track one or two numbers at each of five stages: acquisition (are the right people installing?), activation (do new users reach the value moment?), retention (do they come back?), revenue (does retained use turn into money?) and quality (is the app stable enough not to lose them?). Define each metric before you look at it, take it from one named source, and compare cohorts with your own history rather than with a published benchmark, because no reliable public benchmark exists.
This page is for founders and product owners who have an app in the stores and a dashboard full of totals. It sets out the framework, explains how Apple, Google and Firebase each define the same words differently, and then covers the tracking plan that feeds it. For what to do once the numbers show a retention problem, see how to improve mobile app retention.
Key takeaways
- Downloads and sessions are totals. They cannot tell you whether the app is working; cohorts can.
- Define activation for your own product. Installing is not activating.
- Apple's retention figures cover opted-in devices only, GA4's retention report covers the first 42 days, and Crashlytics counts installations. The tools will disagree, and that is expected.
- Google publishes numeric quality thresholds: user-perceived crash rate 1.09% and ANR rate 0.47% overall, 8% per phone model.
- No retention benchmark with a published method exists. Compare against your own earlier cohorts.
The founder measurement framework
Read the table from left to right for each stage: the question it answers, a working definition, where the number comes from, and the caveat that most often leads founders astray. Start with one metric per row. Add more only when a decision needs them.
| Stage | Question | Definition | Data source | Caveat |
|---|---|---|---|---|
| Acquisition | Are the right people arriving? | Store listing visitors who go on to install (conversion), and installs by source | App Store Connect Analytics; Play Console store listing reports | On iOS, user-level attribution needs ATT consent; expect aggregated data from Apple's privacy-preserving attribution instead |
| Activation | Do new users reach the value moment? | Share of new users who complete your defined activation event within N days of first open | Your own events in Firebase / GA4 or a product analytics tool | Activation must be defined per product. If the event is wrong, every downstream number is misleading |
| Retention | Do they come back? | Share of a cohort that returns on day 1, 7 and 30, or each week, chosen to match how often the job naturally recurs | GA4 retention overview and cohort exploration; App Store Connect retention | Apple counts only opted-in devices that installed and then opened the app. GA4's retention report covers the first 42 days. A weekly-use app judged on daily return will look broken when it is not |
| Revenue | Does retained use turn into money? | Trial to paid, renewals, revenue per user by cohort, refunds | Store subscription and sales reports; server-side purchase events | Verify purchases on the server. Client-side events can be lost or duplicated |
| Quality | Is the app stable enough not to lose users? | Crash-free users (Crashlytics); user-perceived crash and ANR rates (Play); hangs and launch time (Xcode Organizer) | Crashlytics, Play Console Android vitals, Xcode Organizer, MetricKit | Each tool uses a different denominator. Crashlytics counts installations and fatal crashes only; Play counts daily users |
Why your tools disagree
Founders often lose a week reconciling numbers that were never meant to match. These are the published definitions behind the most common disagreements.
| Tool | What it actually counts |
|---|---|
| App Store Connect Analytics | Usage data comes only from users who have opted in to share analytics with developers. A session is use for at least two seconds. Retention is the percentage of active devices that installed on a given day and opened the app a set number of days later; installs never opened are excluded, and cells can be blank when privacy thresholds are not met. |
| Google Analytics 4 | The retention overview shows the percentage of users who return each day in their first 42 days, with day 1, 7 and 30 cohort lines. 'New' and 'returning' depend on GA4's user identifier, not on a person. |
| Firebase Crashlytics (vendor documentation) | Crash-free users is 1 minus crashed users divided by all users, where a user is an individual installation of the app on a device. Only fatal events count. It is an aggregation over the selected period, not an average of daily figures. |
| Play Console Android vitals | Crash and ANR rates are normalised by daily users (one user on one device on one day) and only count problems while the app was in active use. Google generally evaluates the last 28 days. |
Pro tip
Pick one source per question and write it next to the metric in your tracking plan. Never divide a number from one tool by a number from another.
Is there a good retention benchmark?
No reliable one. We looked for a primary source that publishes app retention benchmarks with a stated method and found none; the percentages quoted in vendor articles do not say which apps, categories, periods or definitions they used. A benchmark built on unknown definitions cannot be compared with your number, which is itself defined by one of the tools above.
What works instead: compare each weekly or monthly signup cohort with your earlier cohorts on the same definition, and look for the point where the curve flattens. A flat line, at whatever level, means a group of users has found lasting value. A curve that keeps falling to zero means they have not, and the fix is in the product, not in the dashboard.
Event Tracking and the Tracking Plan
Start from questions, not from everything that can be tracked. Define a tracking plan: event names in a consistent format (for example object_action, like order_completed), the properties each carries, when it fires, and which framework stage it serves. Document it and review it when features change. The same evidence-first discipline underpins ZSpace Labs' conversion optimisation work.
Properties such as plan, app version, platform, acquisition source and signup date let you segment events. Keep them non-sensitive; don't send names, emails or precise location unless there's a clear, disclosed purpose and a lawful basis.
Activation and Onboarding
Define the activation event, then track each step from first open to that event. This is the most useful early funnel for most apps, and the onboarding guide explains how to improve it. A good activation event is something a retained user almost always did early and a churned user often did not; treat that link as a hypothesis to test, not proof of cause.
Retention, Cohorts and Funnels
Measure retention by cohort (for example, users who signed up in a given week) so changes to the product show up as differences between cohorts. Choose the retention definition that matches your app's natural frequency; a weekly-use app shouldn't be judged on daily return. GA4's cohort exploration lets you choose both the inclusion criterion and the return criterion, so you can count a return as a meaningful action rather than any app open.
Funnels show where users drop off in multi-step flows like signup, checkout or booking. Session data (length, frequency, screens viewed) adds context but can mislead on its own: a short session may mean the app did its job quickly.
Revenue Metrics
For commerce and subscription apps, track purchases, revenue, refunds, trial starts and conversions, and subscription renewals and cancellations, by cohort. Verify revenue events on the server, since client-side events can be lost or duplicated; see mobile app payments. Which revenue metrics matter depends on the model; see how to monetise a mobile app.
Quality Metrics and Google's Thresholds
Stability and speed affect every other stage. Google Play's Android vitals is the only official source we found that publishes numeric quality bars. Its bad-behaviour thresholds are a user-perceived crash rate of 1.09% overall and 8% per phone model, and a user-perceived ANR (Application Not Responding) rate of 0.47% overall and 8% per phone model. Apple publishes no equivalent threshold.
Watch these per release and per device model, and correlate releases with changes in activation and retention. For reading crash and ANR data and finding the cause, see the founder's diagnostic guide to crashes and slow apps.
Attribution and App Tracking Transparency
Attribution links installs and conversions to campaigns. On iOS, iPadOS and tvOS 14.5 and later, Apple requires permission through App Tracking Transparency to track users or access the device's advertising identifier. Apple defines tracking as linking user or device data from your app with other companies' data for targeted advertising or advertising measurement, or sharing it with data brokers. For measurement without that consent, Apple offers AdAttributionKit, which reports in aggregate. On Android, the Play Install Referrer supplies install source information. Expect aggregated, less granular data than in the past.
Push Notification Analytics
Measure notifications by what happens next, not just opens: did the user complete the action the message was about, and did opt-outs rise? See push notifications.
Privacy-Aware Analytics
Collect what answers your questions and nothing more, avoid personal data in events, respect consent where required, disclose analytics accurately in store privacy declarations, and set retention periods. The mobile app data privacy guide covers the details.
A monthly founder review: five numbers, one decision
Once a month, put one number from each framework stage on a single page, each with its source and the same figure for the previous three cohorts. Then write down one decision the numbers support: what you will change, which cohort should show the effect, and when you will look again. If no decision follows, the metric is probably not the right one for this stage of the product.
- Acquisition: listing-to-install conversion for your main source
- Activation: share of last month's new users who reached the activation event
- Retention: week 4 (or day 30) return for the latest complete cohort, against the previous three
- Revenue: trial-to-paid or first purchase rate for the same cohort
- Quality: user-perceived crash and ANR rate on Android, and crash-free users by app version
- One decision, with the cohort and date you will check it
What to Track at Each Stage
| Stage | Focus |
|---|---|
| MVP / early launch | Activation funnel, crashes, a few core actions, qualitative feedback |
| Finding product fit | Cohort retention, feature adoption, where engaged users differ |
| Growth | Acquisition channels, conversion funnels, revenue per user |
| Mature product | Experiments, long-term retention, churn signals, efficiency |
Hypothetical example: three dashboards, three answers
Hypothetical example: a founder of a meal-planning app sees healthy downloads but cannot tell whether week 1 is going well. App Store Connect shows one retention figure, GA4 another, and the agency's Crashlytics report says stability is fine. Using the framework, the team defines activation as 'saved a first weekly plan within three days', takes retention from GA4's cohort exploration with that event as the return criterion, and ignores the App Store Connect retention figure for this question because it covers only opted-in devices. The first clear finding is that users who skip saving a plan rarely return in week 2, which points to onboarding rather than to push campaigns. The example is illustrative and does not describe a real client.
Limits of This Framework
The framework suits consumer and prosumer apps where individual usage drives value. For B2B apps where an organisation buys seats, account-level measures such as active seats and renewals matter more than individual day-30 return; reducing SaaS churn covers that model. Very small apps may find that App Store Connect shows blank cells or that Organizer lacks data, because privacy thresholds are not met; qualitative interviews then carry more weight than any dashboard. And metrics show where something is wrong, not why. Pair them with user research before committing to a fix.
Common Analytics Mistakes
- Tracking everything with no plan or documentation
- Inconsistent event names across platforms
- Personal data in event properties
- Relying on client-only revenue events
- Focusing on downloads instead of activation and retention
- Comparing to generic benchmarks instead of your own cohorts
- Dividing a number from one tool by a number from another
- Never validating that events fire correctly after releases
Conclusion and Next Steps
Good app analytics starts with questions and a documented tracking plan, follows the path from acquisition to quality, analyses cohorts rather than totals, and respects privacy. This month, write one definition and one source for each of the five stages, then hold the first five-numbers review. When retention is the weak stage, move to retention diagnosis and levers; when quality is, use the crash diagnostic guide.
For the wider build context, see the mobile app development guide. For web and SaaS products, see product analytics implementation. For measuring the store listing itself, see app store optimisation.
Want a tracking plan you can trust?
ZSpace Labs is a remote-first technology studio that works with clients globally on mobile app development. If useful, we can talk through an analytics and tracking plan tied to the decisions you need to make.
Sources
Apple: App Store Connect Analytics: metrics definitions; App Store Connect Analytics: app retention; User privacy and data use (App Tracking Transparency).
Google: Google Analytics Help: Retention overview report; Google Analytics Help: cohort exploration; Android vitals (updated 9 October 2026); Play Console Help: monitor technical quality with Android vitals; Firebase Crashlytics: crash-free metrics (vendor documentation).
Accessed 15 October 2026. Tool definitions and report names change; check the current documentation before relying on a figure.
Common questions.
Collecting and analysing data about how people use an app, such as which features they use, where they drop off and whether they return, to make product decisions.
The key steps of your core flows: signup or first open, onboarding steps, the activation action, the main value actions, purchases or conversions, and errors that block users.
The point where a new user first experiences the app's core value, such as sending a first message or completing a first order. You define it for your own product; an install is not activation.
As the share of a cohort of new users who return after a set period, such as day 1, day 7 or day 30, or who perform a key action in later weeks. Each tool defines it slightly differently: Apple counts only opted-in devices that installed and then opened the app, and GA4's retention report covers a user's first 42 days.
There is no reliable public benchmark. We found no primary source that publishes app retention benchmarks with a stated method, and the figures in vendor blogs do not say how they were calculated. Compare each new cohort with your own earlier cohorts and watch the trend.
Attributes describing a user or device, such as plan type, app version or signup date, used to segment events. Avoid storing sensitive personal data as properties.
Grouping users by a shared characteristic, most often the week they signed up, and comparing how each group behaves over time.
On iOS, iPadOS and tvOS 14.5 and later, apps need permission through ATT to track users or access the advertising identifier. Apple defines tracking as linking data from your app with other companies' data for targeted advertising or ad measurement, or sharing it with data brokers. First-party product analytics inside your own app is still possible, but must be disclosed accurately.
Tracking everything without a plan, inconsistent event names, no documentation, sending personal data in events, and only looking at vanity metrics like downloads.
They count different things. App Store Connect usage data comes only from users who opted in to share analytics, Crashlytics counts app installations rather than people, and Android vitals normalises by daily users over roughly the last 28 days. Pick one source per question and do not mix them in a single ratio.