> ## Documentation Index > Fetch the complete documentation index at: https://documentation.onesignal.com/llms.txt > Use this file to discover all available pages before exploring further. # Interpreting retention curves > Read and act on OneSignal retention curves. Covers a worked cohort example, what flat and steep curve shapes tell you about return behavior, how to compare channels and platforms, and the methodology behind the numbers. Read the [Retention](./retention-metrics) chart to understand how often your users come back and what its curve shapes tell you about your audience. ## What retention answers Retention metrics tells you whether active users continue to come back to your app. Pick a day, and the chart follows the people who were active then to see how many return on the days that follow. Many things influence whether someone returns, including your messaging, so treat a change in the curve as a signal to investigate rather than a measure of what caused it. When you want to connect specific messages to what users did next, use [Conversion metrics](./conversion-metrics), which attribute conversions to the messages that drove them. ## A worked example Follow one cohort through five days to see how the numbers come together. A cohort is a group of people you track over time. Say 1,000 users opened your app on July 28. Those 1,000 users become the cohort, and the group stays fixed from that point on. Nobody joins it later and nobody drops out of it. The group holds together even if their Subscriptions change during the window. From there, the chart looks at each following day and counts how many of those same 1,000 users came back. | Day | Date | Cohort members active | Retention rate | | :---- | :------- | :-------------------- | :------------- | | Day 0 | July 28 | 1,000 | 100% | | Day 1 | July 29 | 420 | 42% | | Day 2 | July 30 | 310 | 31% | | Day 3 | July 31 | 335 | 33.5% | | Day 4 | August 1 | 300 | 30% | Day 3 is higher than Day 2, which can be a normal fluctuation of this chart. OneSignal measures each day independently against the original 1,000, so someone who skipped July 30 and returned on July 31 counts on Day 3. The curve tracks how much of the original group is active on each day, so it can rise as well as fall. ## Reading the shape Look at the shape of the curve before you read any single number. **Falls, then flattens.** Part of the cohort stops returning over the first few days, and the rest keep coming back at a steadier rate. The level where it flattens tells you roughly how much of that day's audience returns regularly. **Stays high and flat.** Most of the cohort returns most days. This can happen when your app is part of something people do daily, or when the cohort was drawn from a day that skewed toward your most active users. **Falls continuously.** The cohort keeps thinning across the whole window. A 30-day window shows whether it levels off later than a shorter window would reveal, and comparing a recent cohort against an older one shows which direction it is moving. **Very low Day 1.** Most of the cohort did not return the next day. Reasons can include onboarding, the expectations set before someone arrives, or a cohort start date that captured a burst of one-time visitors. Comparing the same shape across different cohort start dates tells you more than any single day's rate, so your own trend over time is the most useful benchmark. ## Comparing channels and platforms Set **Compare by** to Channel or Platform to break the cohort into series. Channel series reflect the channels each user is subscribed to on the cohort start date. They overlap, because a user subscribed to both push and email appears in both series, so do not expect them to add up to the unfiltered total. Read each series against the unfiltered curve rather than against each other. The series for users with no subscribed channels is a useful baseline. Comparing it against the push series shows how differently your subscribed users and your users with no subscribed channels behave, which is the closest the chart comes to showing the effect of being reachable. Platform series divide the cohort by device type. Large gaps between platforms can point at differences in the app itself on each platform rather than in your messaging. ## How retention is calculated These rules determine every number on the chart. **Cohorts are snapshots.** OneSignal holds every attribute of the cohort constant from the cohort start date. Someone who unsubscribes from email on Day 5 stays in the email series for the whole window, because they were subscribed on Day 0. This is what makes cohorts comparable over time. **Days are counted independently.** OneSignal reports whether each cohort member had a session on each day, regardless of the days in between. **The curve ends at the last complete day.** The chart hides incomplete days and draws no projection. A dashed or extrapolated line would suggest a forecast, so the line stops where the measured data ends. **Retention counts users.** A session on any one of a user's devices counts as activity for that user, so someone who uses your app on a phone and a laptop counts once rather than twice. See [Sessions](./sessions) for how OneSignal tracks session data per Subscription and aggregates it at the user level, and [Users and Subscriptions](./subscriptions) for how the two relate. ## FAQ ### My curve goes up on some days. Is that a bug? No. OneSignal measures each day independently against the original cohort, so a user who skips a day and returns later counts on the day they return. A curve that rises on some days is expected. ### Which window should I use? Start with 30 days to see whether the curve flattens, then use 7 or 14 days when you want a closer look at the first few days after the cohort start date. Shorter windows show the early drop-off in more detail. ### Can I compare two cohorts side by side? Not in the chart today. Select one cohort start date at a time and compare the curves yourself, or export each to CSV. ## Related Filters, data availability, and export for the Retention chart. How OneSignal defines a session and aggregates session data across a user's devices. How Users and Subscriptions relate, and why retention counts users. Canonical definitions for every metric across the dashboard, API, CSV, and Event Streams.