Retention by segment: a cohort analysis case

Maya

We're planning next year. The CEO says trip users “churn like crazy” and wants to drop the summer travel campaign. Theo wants to double it. Is our retention healthy for couples, flats and trips?

Alex Tutor

Here's a full year for last summer's sign-ups, by the type of their first group. Two ways of counting activity, plus how many of the active users were in a group created after they signed up.

Theo

Before anyone asks: yes, trips are the biggest group. More than half of last summer's sign-ups came in with a trip, many of them through my travel campaign.

Alex Tutor

Dataset · 39 rows

Halves summer 2025 sign-ups: monthly activity by group type

Users who signed up July–September 2025, followed for 12 months. Month 0 is the sign-up month. Staff excluded.

Download CSV

Columns
group_type: Type of the user's first group: couple, flat or trip
month: Months since sign-up
cohort_users: Users in the segment's cohort
active_opened_app: Users who opened the app at least once that month
active_any_event: Users with any event that month, including recurring bills created by the server
active_in_new_group: Of those who opened the app, users active in a group created after their sign-up month

Data

group_type	month	cohort_users	active_opened_app	active_any_event	active_in_new_group
couple	0	600	600	600	0
couple	1	600	433	439	6
couple	2	600	387	393	6
couple	3	600	362	374	6
couple	4	600	344	356	6

Show 34 more rows

couple	5	600	335	347	12
couple	6	600	330	342	12
couple	7	600	332	344	12
couple	8	600	323	335	12
couple	9	600	323	335	12
couple	10	600	322	334	12
couple	11	600	315	327	12
couple	12	600	316	328	12
flat	0	800	800	800	0
flat	1	800	527	551	8
flat	2	800	439	479	16
flat	3	800	391	447	16
flat	4	800	360	424	24
flat	5	800	344	416	24
flat	6	800	333	413	24
flat	7	800	325	405	32
flat	8	800	327	407	32
flat	9	800	323	411	32
flat	10	800	320	408	40
flat	11	800	317	405	40
flat	12	800	315	403	40
trip	0	1600	1600	1600	0
trip	1	1600	536	536	16
trip	2	1600	198	198	32
trip	3	1600	135	135	48
trip	4	1600	119	119	48
trip	5	1600	89	89	48
trip	6	1600	97	97	48
trip	7	1600	116	116	64
trip	8	1600	121	121	80
trip	9	1600	165	165	112
trip	10	1600	222	222	176
trip	11	1600	310	310	256
trip	12	1600	349	349	304

Start with the measure. We count monthly activity by app opens and use “any event” only as a check. Then read each segment on its own; never average them together.

You

Couples first. 600 users; month 12 is the last row for couples.

Alex Tutor

Question 1

What share of the couples cohort opened the app in month 12? In %, one decimal.

Your answer

Units: % (e.g. 45.5)

52.7% a year later, and flat since month 6. Barely any of them needed a new group: couples keep using the one they started with.

Maya

So couples are the steady part. I suspected as much, but it's good to see it flat on a chart.

You

Flats in month 12: 315 opened the app, but 403 had some event.

Alex Tutor

The difference is people whose only activity is a recurring rent bill the server keeps adding. How many are there, and what share of the cohort?

Alex Tutor

Question 2

Flats, month 12: how many users count as active only because of server-created events (any event minus opened the app)?

Your answer

Units: users

Maya

88 people, 11% of flats. Are they retained or not?

Alex Tutor

We can't tell from this table. Maybe they settle up outside the app, maybe they moved out and nobody switched the bill off. Report 39.4% and flag the 88 as a question worth a look.

You

So the headline for flats is 39.4%, with a footnote about the 88 and a check to run.

Maya

Now the CEO's point. Trips: 1,600 users, and by month 5 only 89 open the app. That's 5.6%.

Alex Tutor

Now read to the end of the rows. Month 12: 349 active again, and look at the last column for them.

You

304 of the 349 are in a new group. They're planning next summer's trip.

Maya

So the CEO's “churn like crazy” is mostly people waiting between trips?

Alex Tutor

Question 3

Trips, month 12: what share of the users who opened the app were active in a new group? In %, one decimal.

Your answer

Units: % (e.g. 45.5)

87.1% of the returning trip users started a new group. The trip group ended with the trip; the users didn't leave Halves, they waited for the next one.

Theo

So monthly retention is the wrong ruler for trips. What should I track instead?

Alex Tutor

Think about the rhythm first. People take one or two trips a year, so a trip metric has to look across a year, not a month.

Alex Tutor

Question 4

Which metric best tracks whether trip users are retained?

Choose one answerShare of trip users active in each month after sign-upShare of trip users who start or join a new trip group within 12 monthsShare of trip users with any event in month 3Average number of expenses per trip group

Theo

Before Maya's memo: the CEO will say trips are our biggest segment with the worst retention, so the travel campaign goes. What do I answer?

Alex Tutor

Answer with the trip rows, and be honest about what one summer's cohort can and can't show.

You

And not just defend the campaign. If trip users only come back once a year, that changes what the campaign is worth.

Alex Tutor

Question 5Short answer · AI-checked task

The CEO says: “Trips are our biggest segment and have the worst retention. Cut the summer travel campaign.” Help Theo answer in 3–4 sentences with numbers from the table, including one honest limit.

Write your answer and get a score with feedback from our AI reviewer.

Log in to get AI feedback

Maya

Now the full memo: for each segment, is retention healthy, by what measure, and what we should do about it.

Alex Tutor

One verdict per segment, each with its own rhythm and its own numbers. And say what you'd check next.

You

Couples, flats with the 88 flagged, trips by return, and no blended curve.

Alex Tutor

Question 6Case · AI-checked task

Write Maya's memo: is Halves retention healthy for couples, flats and trips?

Halves is planning next year's product and marketing priorities. The CEO wants to cut the summer travel campaign because trip users “churn”; Theo wants to double it. Maya asks for a memo based on the table above (summer 2025 sign-ups, 12 months).

Write 200–350 words:

The measure you use and why (and how you treat “any event”).
A verdict for each segment (couples, flats, trips) with at least two numbers each.
What the flats' two curves tell you and what you'd check.
How trips should be judged, and what that means for the campaign debate.
The main limit of this data and what you'd look at next.

Dataset for this task: Halves summer 2025 sign-ups: monthly activity by group type

Write your answer and get a score with feedback from our AI reviewer.

Log in to get AI feedback

Alex Tutor

Three segments, three rhythms, three verdicts. The CEO and Theo were each right about one line of the table.

Alex Tutor

That’s the lesson. You answered every task — nicely done.

Next lesson: Hypothesis, metric and decision rule