Day-N vs rolling vs bracket retention, with examples

Theo

I read that good consumer apps keep at least 25% of users on day 7. Our day-7 retention is 20%. Should I stop buying installs?

Maya

Odd. Our investor deck says week-1 retention is 70%. Same users.

Alex Tutor

Both can be right, because “retention” is three different measurements sharing one word. Let's compute all three on ten users, so you can see each person.

Each row lists the days after sign-up on which the user opened Halves. Day 0 is the sign-up day. We've watched them for 40 days.

Alex Tutor

Dataset · 10 rows

Ten Halves users and their active days

Days since sign-up with `app_opened`, separated by `;`. Observed to day 40.

Download CSV

Columns
user_id: User id
active_days: Days after sign-up on which the user opened the app

Data

user_id	active_days
r01	0;1;7;14;21;28
r02	0;3;9;16;30
r03	0;2
r04	0;7
r05	0;12;33

Show 5 more rows

r06	0;5;8;40
r07	0
r08	0;6;13;20;27;34
r09	0;31
r10	0;10;11

First, day-N retention: the share of users active exactly on day N. It's what Theo's blog means by “day 7”.

Alex Tutor

Question 1

What is the day-7 retention of these ten users (active exactly on day 7)? In %.

Your answer

Units: %

Alex Tutor

Only r01 and r04 opened the app on day 7 itself. But r02 came back on day 9 and r05 on day 12. For a bill-splitting app, that's perfectly normal.

Rolling retention counts a user as retained on day N if they were active on day N or any later day. It answers “who hasn't left yet?”

You

So r09, who opened the app on day 0 and then only on day 31, counts as retained on day 7?

Alex Tutor

Yes. That's generous by design, and it's why rolling is always at least as high as day-N.

Alex Tutor

Question 2

What is the rolling day-7 retention (active on day 7 or any later day)? In %.

Your answer

Units: %

Alex Tutor

The third way is bracket retention: the share active at least once within a window, such as days 7–13, the second week.

You

So r02 on day 9 and r05 on day 12 count, but r09's day 31 doesn't.

Alex Tutor

Exactly. It forgives the exact day but still asks whether the person showed up in that period.

Alex Tutor

Question 3

What is the bracket retention for days 7–13 (active at least once in the second week)? In %.

Your answer

Units: %

Alex Tutor

Same ten people: 20% on day 7, 70% in week 2, 80% rolling. Theo's blog used day-N, the deck used a bracket.

Pick the measure by the natural usage rhythm. Day-N suits products people need daily, like games or chat. For Halves, weekly brackets suit couples and flatmates, and monthly brackets suit monthly reports and rent cycles.

Maya

And rolling? It gives the prettiest number.

Alex Tutor

It's useful for one question: has this user gone for good? As a report it flatters, and it keeps changing after you publish it.

Alex Tutor

Question 4

Match each question to the retention measure that fits it best.

Pick an answer for each row. Choosing an answer that is already used moves it to this row.

A daily puzzle game: do players come back the next day?
Choose an answer
Halves couples: are they still splitting costs in their second week?
Choose an answer
Halves flatmates who log rent: are they back the next month?
Choose an answer
Has this user left for good, or might they still come back?
Choose an answer

You

What do you mean, rolling keeps changing?

Alex Tutor

Compute rolling day-28 retention on day 35: r01, r02, r05, r08 and r09 have been back since day 28. That's 5 of 10, 50%. Then on day 40 r06 opens the app, and the same metric becomes 60%.

Bracket and day-N retention are final once the window has passed. Rolling retention of last spring's cohort can still move next winter.

Alex Tutor

Question 5

True or false: rolling day-28 retention reported on day 35 (50%) is final and won't change later.

TrueFalse

Theo

Fine. So what do I tell my team about that blog post?

Alex Tutor

That the benchmark measures a different thing, and what you'll measure instead. With the numbers from these ten users.

You

And the name of the measure next to every number, so nobody compares 20% with 70% again.

Alex Tutor

Question 6Short answer · AI-checked task

Write Theo 3–4 sentences for his team: why the “25% on day 7” benchmark doesn't show that Halves is failing, and which retention measure Halves should report instead. Use the numbers from the ten users.

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

Log in to get AI feedback

Alex Tutor

“What's our retention?” has no answer until someone says which kind, and for which window.

Alex Tutor

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

Next lesson: Reading a cohort table