Article · Updated August 2026

Why every tool reports a different retention rate

Two retention rates can differ because they retain different entities after different starting events on different time boundaries. App retention, product retention and subscription retention are not interchangeable.

Before comparing percentages, align five fields:

1. cohort entry;
2. entity;
3. return event;
4. time boundary;
5. cohort maturity and exclusions.

Three common retention measures

App retention

Apple defines app retention as the percentage of active devices that installed on a selected day and opened the app on a later exact day. Devices that install but never open do not qualify; usage is based on people who opted in to share analytics, and privacy thresholds can leave cells blank (Apple).

Entity: active device. Entry: install plus eventual first open. Return: open on an exact day after install.

Product retention

A product analytics tool may define a cohort from an account’s first-value event and retain it when the account repeats a chosen event in a later day or week. Depending on configuration, it can measure users, accounts or anonymous devices and use exact, rolling or unbounded return windows.

Entity and events are chosen by the team. That flexibility is useful and makes the label essential.

Subscription retention

RevenueCat’s subscription retention cohorts start with a paid subscription and measure successful renewal through later billing periods. It calculates relative to the original cohort members that had an opportunity to renew and excludes incomplete periods from its relative total (RevenueCat).

Entity: subscription. Entry: first paid standard period. Return: additional payment at the relevant subscription period.

Why the numbers differ

Build a retention definition sheet

For every chart, record:

Once named this way, the rates no longer “disagree.” They describe different retention.

A fictional reconciliation

Acme Coach sees:

• App Store Connect D7 app retention: 22%;
• product analytics D7 return: 54%;
• RevenueCat month-1 subscription retention: 81%.

These cannot be placed in one comparison column.

The App Store denominator is eligible active devices from the install cohort; product analytics uses identified accounts that completed signup; RevenueCat begins after first paid purchase. Every later denominator is more selected.

Create a bridge:

Then retain each stage on its own terms. Do not call 81% subscription retention evidence that 81% of installers remain.

Diagnose suspicious mismatches

Check the definition first

Find the vendor’s metric documentation and your saved chart configuration. Do not compare dashboard labels alone.

Check identity

Look for anonymous users becoming identified, account merges, test devices, reinstall behaviour, shared accounts and multiple subscriptions.

Check time

Align timezone, exact-day versus rolling periods, cohort start and maturity. A day-boundary mismatch can move returns between cells.

Check lifecycle rules

RevenueCat documents differences with App Store Connect around billing retry: RevenueCat can credit a later payment from the same subscriber/product as renewal, while App Store Connect may treat payments during billing retry or later as new subscriptions. That can change both starts and retention (RevenueCat).

Check environment and filters

Separate sandbox/test, app version, product, store, country and subscription duration. An unreleased app showing activity can indicate test or sandbox data rather than real customers.

Publish comparable measures

Use full names:

• D7 active-device app-open retention;
• Week-1 activated-account first-value retention;
• Month-1 paid-subscription renewal retention;
• Month-1 retained recurring revenue.

Include the cohort window, sample size and incomplete-period treatment. Link back to the source chart.

Reproduce a rate from event data

When the source permits export, reproduce one cohort before deciding the dashboard is wrong.

For exact D7 product retention:

1. select accounts whose entry event occurred on the cohort date;
2. exclude the documented internal/test accounts;
3. apply the reporting timezone;
4. count each eligible account once;
5. count accounts with the return event on calendar day 7;
6. divide retained accounts by the eligible cohort;
7. leave a recent cohort incomplete until day 7 closes.

For weekly retention, define whether the following week is a calendar week, days 7–13, or any return after day 7. Those produce different rates.

Exact, classic and rolling retention

Rolling retention will generally be at least as high as exact retention for the same event and cohort because it accepts more return times. Do not compare them under the same D7 label.

Customer, logo and revenue retention

SaaS teams also use “retention” for commercial measures:

Stripe notes that revenue retention can exceed 100% because expansion can increase the remaining MRR from a cohort (Stripe). That is not the same as more than 100% of customers being retained.

A step-by-step mismatch audit

When two charts disagree, fill this sequence without skipping ahead:

1. Copy the full metric name and source URL.
2. Record filters and date range.
3. Record entry, entity, return and boundary.
4. Record consent, privacy and test-data exclusions.
5. Mark incomplete cells.
6. Compare raw cohort sizes before rates.
7. Reproduce one cohort from export where practical.
8. Rename the measures if definitions differ.
9. Escalate as a data defect only if identical definitions still produce unexplained results.

Raw cohort size is often the fastest clue. If Apple begins with 400 active devices and product analytics begins with 250 identified accounts, a percentage comparison cannot resolve the missing 150 entities.

Which measure belongs in the weekly report?

Choose the rate tied to the current constraint:

• acquisition quality: retained first-value accounts by source;
• product habit: repeated core-value event by account;
• subscription health: successful renewals by paid cohort;
• revenue durability: gross or net retained recurring revenue.

Keep the full name and definition visible. The founder report can show one rate and link to the other source views without declaring one tool the winner.

Frequently asked questions

Sources

• Apple app retention definition
• RevenueCat subscription retention
• Stripe Billing analytics and retention definitions
• PostHog versus Apple D7 discussion
• RevenueCat versus App Store discussion
• Apple Analytics versus RevenueCat discussion