Customer Retention Analysis SOP

A customer retention analysis shows whether people keep using your product after the first excitement wears off, and it is the most honest product-market fit test you can run. You group users into cohorts by start date, track how many return each week or month, and look at the shape of the curve. Setup takes two or three days; a first reading needs eight to twelve weeks of data. Surveys tell you what people say about your product, and this SOP tells you what they do.

What is customer retention analysis and when should you use it?

Retention analysis measures the share of a starting group of users who are still active after a set time. It works with cohorts: everyone who signed up in March is one cohort, everyone from April is the next. Comparing cohorts shows whether your product improves over time and removes the noise of a marketing spike.

Growth writers including Andrew Chen and Brian Balfour describe the flattening retention curve as the clearest behavioural sign of product-market fit. A curve that drops and then levels off means a group of users found lasting value. A curve that keeps sliding toward zero means you are filling a leaky bucket.

Use this SOP:

• After a Sean Ellis test, to check that the users who say they would miss you also stay.
• Before spending on paid acquisition, because new users poured into a leaking product are wasted money.
• When growth looks good but revenue doesn't follow, to see whether new signups hide old users leaving.
• After a big product change, to compare cohorts that joined before and after it.

What do you need before you start?

• Event tracking with a user ID and timestamps. PostHog, Mixpanel and Amplitude all have free tiers with cohort reports built in. The digital behavior tracking SOP covers the setup.
• At least three cohorts with 50+ users each, so one unusual week doesn't decide the outcome.
• A definition of "active" tied to value, written down before you look at the data.
• Billing data if you sell subscriptions, to measure revenue retention next to user retention.

How do you run it, step by step?

1. Define the action that counts as retained

Logging in is a weak signal. Pick the action that delivers the value you promise: "sent an invoice", "booked a class", "completed a lesson". A user who opens the app and leaves hasn't received any value, and counting them inflates the curve.

2. Pick the natural interval

Match the interval to how frequently the problem occurs. A messaging app needs daily or weekly retention; a payroll tool runs monthly; a tax tool runs yearly. Measuring a monthly product weekly makes healthy users look like they left.

3. Build the cohort table

Rows are cohorts by start week or month. Columns are periods since start (period 0, 1, 2 and onward). Each cell is the share of the cohort that did the value action in that period. Most analytics tools generate this table in a few clicks; a spreadsheet works for under 1,000 users.

4. Plot the curves and read the shape

Draw one line per cohort. Look for three things: where the steep early drop ends, whether the line levels off, and the level where it flattens. A curve that flattens at 25% means one in four users found a lasting habit, and that group is your real market.

5. Compare newer and older cohorts

If newer cohorts sit above older ones, your product changes are working. If newer cohorts sit below, look at what changed: a new acquisition channel bringing weaker-fit users, a pricing change, or a worse first-week experience.

6. Split curves by segment

Recalculate retention by acquisition channel, customer type, plan and the features used in week one. One segment holding at 50% while the rest drops to 5% tells you exactly who to build and market for.

7. Ask the people who left

Email ten to fifteen users who dropped off after period 1 and ask for a 15-minute call about what they were trying to do. Retention data shows where people leave; only conversations show why.

Template you can copy

RETENTION ANALYSIS SHEET

Definitions (fill in before opening the data)
Value action:        [e.g. "created and sent an invoice"]
Interval:            [daily / weekly / monthly]
Cohort size minimum: [50 users]

Cohort table (share of cohort doing the value action)
Cohort   | Users | P0   | P1  | P2  | P3  | P4  | P5  | P6
Jan      |       | 100% |     |     |     |     |     |
Feb      |       | 100% |     |     |     |     |     |
Mar      |       | 100% |     |     |     |     |     |

Segment table (retention at period 3)
Segment / channel | Users | P3 retention | Note

Revenue retention (subscriptions)
Gross revenue retention = (starting MRR minus churn and downgrades)
                          / starting MRR
Net revenue retention   = (starting MRR minus churn and downgrades
                          plus expansion) / starting MRR

Exit call notes
User | Last active | What they hired the product for | Why they stopped

Worked example: a meal-planning app

This example is illustrative. A meal-planning app tracks "planned a week of meals" as its value action and uses weekly retention. Four monthly cohorts of 400 to 600 users each show this at week 8:

• January cohort: 9%, still falling.
• February cohort: 11%, still falling.
• March cohort, after a grocery-list feature shipped: 17%, flat since week 5.
• April cohort: 19%, flat since week 4.

The founder splits the April cohort by household type. Families with children hold at 34%, single users at 6%. Exit calls with single users reveal that planning seven dinners for one person feels like overkill. She keeps the family focus, shifts her ads to parenting communities, and runs a product usage analysis to see which features the 34% use every week.

What should you measure, and when do you stop?

Stop the analysis once three consecutive cohorts give the same shape: the answer won't change with more data. If no segment flattens after three product changes, revisit the problem with the problem statement validation SOP before building more features.

Which mistakes should you avoid?

• Counting logins as retention. Tie "active" to the value action or the curve flatters you.
• Using the wrong interval. A monthly product measured daily looks broken.
• Reading a blended average. Total active users can grow while every cohort shrinks. Always look at cohorts.
• Judging from one cohort. A launch-week cohort full of curious visitors behaves differently from steady signups.
• Fixing acquisition first. Better ads into a sliding curve only speed up the leak.

How can AI speed this up?

Export the cohort table as CSV and ask an AI assistant to plot the curves, flag where each one flattens and compare segments. It can also cluster exit-call notes into reasons for leaving. Check two things by hand: that the AI used your value action and not raw logins, and that small segments (under 50 users) aren't treated as proof. An AI will happily report a 60% curve built on twelve people.

The F/MS AI for startups module shows founders how to run this kind of analysis with AI tools without a data team.

Where does this SOP fit in your validation plan?

Retention analysis is the behavioural backbone of the Product-Market Fit stage. Before it, run the Sean Ellis test for a quick stated signal. Next steps:

1. Find which features the retained users rely on with product usage analysis.
2. Track whether retained customers reach their goals with customer success metrics.
3. Place retention in the full funnel with the AARRR metrics framework.
4. Turn retention into money with customer lifetime value analysis.

See all Product-Market Fit SOPs. The F/MS startup playbook blog has more guides on growth and fit for first-time founders.

FAQ

What is a good customer retention rate for a startup?

The shape matters more than a fixed number. A curve that flattens and stays flat shows fit; the level it settles at depends on the category, with B2B subscriptions settling far higher than consumer apps.

How do I calculate cohort retention?

Divide the number of users from a cohort who did the value action in a period by the cohort's starting size. Repeat for each period and each cohort.

How long should I wait before analysing retention?

Wait until you have at least three cohorts with several periods each. For a weekly product that means about eight weeks; for a monthly product, four to six months.

What is the difference between churn and retention?

Retention is the share of a cohort that stays active; churn is the share that leaves. For one period, churn equals 100% minus retention.