Retention After a Launch Spike: Who Stays?
Retention · 10 min read ·
A launch brings a crowd. Retention shows how many stay. How to read cohort tables, set a fair definition of active, and tell curiosity from real adoption.
A launch day graph is a thrilling thing. A line leaps up, a counter spins, messages arrive. The question that matters comes later and quieter: how many of those people are still here next month?
That question is retention, and it is the measure that separates a spike from a product. This guide explains how to measure it, how to read it after a launch, how to tell curiosity from adoption and what to do with what you find.
What retention means
Customer retention, in the plainest sense, is the ability of a business to keep its customers over time. For a product, we measure it as the share of people who started using it who are still using it after some period.
Two choices define the number.
Who started. Everyone who signed up? Everyone who completed set-up? Everyone who tried the main feature? The wider your starting group, the lower the percentage.
What counts as still using. Logged in? Completed a core task? Paid? A stricter definition gives lower, more meaningful numbers.
Write both choices down. Changing them later without saying so makes comparisons meaningless.
Why cohorts
Overall averages blur everything. If you gain new users each week, the average retention of "all users" mixes people at very different stages. A cohort fixes this by grouping users by when they started.
Cohort analysis tracks groups of users who share a characteristic, usually their start date, and compares how each group behaves over time. A cohort table has one row per start period and one column per period since starting. It shows, at a glance, how each week's new users behaved in their first week, second week and so on.
The launch-week cohort is one row. The week-two cohort is the next. Comparing rows tells you whether newer users stay longer or shorter than the launch crowd.
Building a simple cohort table
You can do this in a spreadsheet.
- Choose the start event: sign-up, say, with the date.
- Choose the active event: completing the core task, with the date.
- For each start week, count how many users started.
- For each following week, count how many of that group were active.
- Divide active by started to get a percentage.
- Arrange the results in a grid: start weeks down, weeks since start across.
Typically, week zero is 100 per cent by definition if you define activity as the start event, or a lower number if you require the core task. The row then fades as people drop away.
Colour the cells in shades so the grid reads like a heat map, and label the numbers anyway. Colour alone is not enough for anyone reading in greyscale or with limited colour vision.
What the launch cohort usually looks like
Launch crowds are mixed. Some come from curiosity, some from friendship, some from a genuine need. A common pattern is:
- A large first row, with many sign-ups.
- A steep early drop, as casual visitors leave.
- A flatter tail, representing those who found real value.
The level where the curve flattens matters most. It is the share who stuck. If the curve keeps falling towards zero, you have not yet found enough value for people to keep.
Do not be alarmed that the launch cohort retains less than later ones. It often does. Later cohorts tend to come from more targeted sources, such as search, referrals and listings that attract people with a specific problem. Comparing cohorts shows whether your product is getting better at keeping people, or whether your audience is getting better matched.
Define active carefully
The word "active" does a lot of work. Consider options.
Logged in. Easy to count, weak evidence. People log in out of habit or by accident.
Opened the app. Similar.
Completed the core task. Stronger. For a notes app, saved a note. For an invoicing tool, sent an invoice.
Used on several days. Shows habit.
Paid. The strongest, for products that charge.
Choose the definition that reflects real value. Report it in words next to the table.
Also consider how often the product is meant to be used. A tool people use daily should be measured weekly or daily. A tool people use monthly, such as a tax assistant, should be measured monthly. Measuring a monthly tool weekly will show false decay.
Read the table in layers
Layer one: the first period. What share did the core task within their first week? If it is low, the problem is early: set-up, clarity or first value. Fix onboarding before you worry about long-term.
Layer two: the drop. Where is the steepest fall? A cliff in week one points to onboarding. A slide in weeks three to six points to missing ongoing value.
Layer three: the plateau. Where does the curve level? That is your base of engaged users.
Layer four: differences between rows. Do later cohorts do better? Which changes did you make between them?
Layer five: the reasons. Numbers say how many, not why. Talk to people who stayed and to people who left.
Separate curiosity from adoption
After a spike, many sign-ups were never going to use the product. To avoid fooling yourself:
- Track by source. Retention from a community post will differ from retention from a niche directory. Combine your referral report with your cohort table.
- Use a stricter starting point, such as people who completed set-up, to see retention among the engaged.
- Look at the time to first value. People who reach value quickly retain better.
- Ask departing users why they left, with a short optional question.
The aim is to understand who your real users are, not to improve a percentage by cherry-picking.
Small numbers
After a modest launch, cohorts are small. A cohort of twenty people has a percentage that jumps in steps of five points when one person leaves.
- Show counts beside percentages.
- Combine adjacent weeks into larger cohorts to see the shape.
- Be cautious about conclusions from single cells.
- Talk to the people.
A pattern that appears across several cohorts is more trustworthy than a single odd cell.
Survivorship and honesty
Retention is partly a story of who is left. Those who stayed are, by definition, the ones who liked the product. If you only ask current users what they think, you learn from survivors. Survivorship bias means focusing on the entries that passed through a selection and overlooking those that did not. Make sure you also hear from those who left.
When sharing retention numbers publicly, be accurate about the definition and the group. "Eight in ten of people who completed set-up were still active after a month" is a different claim from "eight in ten users stay". Advertising rules in many places expect objective claims to be supportable, so keep the evidence.
What improves retention
Retention is not a number to manipulate. It follows value. Typical levers:
- Faster first value. Cut steps between sign-up and the core task.
- Clear next steps. Guide people to the second and third use.
- Reminders that help, not nag, sent with consent.
- Fixing the reasons people leave, from your interviews.
- Fit. Attract the right people in the first place, through honest descriptions and well-chosen listings.
- Genuine improvements. Features people asked for, shipped and announced.
- Support. Fast, kind replies to problems.
Change one thing at a time, and note it in your log.
Setting a sensible expectation
There is no universal good retention figure. It depends on what the product is, how often it should be used and who it is for. A consumer game, a business tool and a seasonal service differ widely. Compare with your own earlier cohorts first. If you compare with others, compare like with like, and treat published figures with caution.
A weekly retention routine
- Add last week's new cohort to the table.
- Update the other rows with this week's activity.
- Look at the shape, not only the latest cell.
- Note any change in onboarding or product that might explain a shift.
- Talk to two users: one who stayed and one who left.
- Write one line on what you will try this week.
Fifteen minutes, once a week, keeps the number connected to the people behind it.
On this site
The launches page shows what visitors see during a launch week, and the leaderboards page shows how community ranking works week by week, separate from anything you can buy. The developers page describes the free API and MCP server for reading listing data into your own sheet, and the blog has further guides in this series.
A short worked example
A product signs up two hundred people in launch week. Forty complete the core task in their first week. Eight weeks later, fourteen of those forty are still completing it weekly. The cohort that signed up in week three is smaller, sixty people, but twenty-five complete the core task in week one and twelve are still active at week eight.
Read plainly: the launch crowd gave the product a wide start and a thin base. The later cohort, arriving through a niche directory and two referrals, stayed in a much larger proportion. The lesson is not that the launch failed. It is that the sources which attract people with a specific problem produce the users who stay, and the next effort should go there. The numbers are small enough that the founder talks to every one of the twelve, which turns percentages back into people.
The short version
Retention is the share who started and are still active after a period. Track cohorts by start week, define active as a real behaviour, read the table for the first-period rate, the drop, the plateau and the differences between rows, separate curiosity from adoption using source and stricter starting points, show counts with percentages and talk to leavers as well as stayers. A spike is a crowd; a plateau is a user base.
Questions and answers
- What is retention?
- The share of people who started using a product who are still using it after a given period, such as a week or a month.
- What is a cohort?
- A group of users who started in the same period, such as launch week, tracked together over time.
- Why is retention lower after a launch spike?
- Spikes bring curious visitors who are less committed. Retention among launch-week sign-ups is often lower than among later, more targeted ones.
- What counts as active?
- A behaviour that shows real use, such as completing a core task. Merely logging in is weaker evidence.
- How many weeks should I follow a cohort?
- Long enough to reach a stable level, often eight to twelve weeks for products with weekly use.