Retention

Week-4 retention is the number we trust most. Here is why.

Day-1 retention flatters onboarding and day-90 arrives too late to act on. After three years of measuring our own product, week 4 is the one number we check before every roadmap review.

DO

Danielle Okafor

Head of Product8 min read

A hand sketching a retention curve on paper beside a coffee cup
Every retention curve we draw starts on paper before it starts in a query.

Most teams we talk to track retention at three points: day 1, day 30 and day 90. Each answers a real question, but none of them tells you early enough whether a feature changed behaviour. Day 1 measures your onboarding email. Day 90 measures a quarter you have already lived through.

Week 4 sits in between. A user who comes back in their fourth week has done it without a welcome sequence, a trial deadline or a sales call nudging them. That return is a habit forming, and habits are what the roadmap is supposed to create.

What we actually measure

We count a user as retained in week 4 when they perform a core action — not a login — between day 22 and day 28 after signup. For Tessaly that action is opening or sharing a saved report.

week_4_retention.sql
select
  date_trunc('week', u.signed_up_at) as cohort,
  count(distinct u.id) as users,
  count(distinct e.user_id) filter (
    where e.occurred_at between u.signed_up_at + interval '21 days'
                            and u.signed_up_at + interval '28 days'
  )::float / count(distinct u.id) as week_4
from users u
left join events e
  on e.user_id = u.id and e.name in ('report_opened', 'report_shared')
group by 1
order by 1;

Three reasons it beats the alternatives

  1. It is early. Four weeks is short enough to read the effect of a release within one planning cycle.
  2. It is honest. Trial reminders and onboarding emails have stopped by day 22, so they cannot inflate it.
  3. It predicts revenue. Across 1,400 workspaces, week-4 retention explains 71% of the variance in month-6 paid conversion.

We stopped debating which features mattered. If week 4 did not move, the feature was not done.

Marcus Hale, Head of Growth at Pellora
A product team sorting feature ideas on a sticky-note wall
Roadmap review at Pellora: every card carries its expected effect on week-4 retention.

Where it misleads

Week 4 is a weak signal for products used monthly, like invoicing or payroll. For those, measure the second cycle instead: did the customer come back for their next invoice run?

None of this needs a data team. In Tessaly, open Retention, set the window to days 22–28 and choose your core action. The chart you get is the one we put at the top of every roadmap review.