Builder Field Guide

Did your first users come back?

How many of your first users came back to do the job a second week? This worksheet gives you the count as well as the percentage, even when some people are still too new to measure.

Week-one retention calculator

Divide the number who returned by the number whose week-one window has closed and whose outcome you know. Here, week one starts seven days after each person’s first qualifying action and ends just before day 14. Use a tally of people. You can keep their names and event logs in your own records.

Your cohort counts

Replace the example with your counts. The calculation runs on your device.

Count each person once, on their first qualifying action.
Include everyone whose first action was less than 14 days ago, even if they already returned.
Count people whose window ended but whose records cannot tell you whether they returned. Enter 0 if all outcomes are known.
Among people with closed windows and known outcomes, count each person who repeated the action at least once during week one.

Observed week-one retention

62.5%

5 of 8 returned in week one.

Still open
3
Closed with unknown outcome
1
Closed with known outcome
8

How much does one return change?

Hold the same 8 people in the denominator.

One fewer return
4/8 · 50%
One more return
6/8 · 75%

Across all 9 closed windows, possible retention runs from 55.6% to 66.7%, depending on who returned during the gaps.

Download these counts (.csv)

Choose the action and the clock

First qualifying action
Choose something a person does to get value. For a shift rota app, that could be publishing a rota to the team. Someone who creates an account and leaves has yet to enter this particular cohort. A signup-based cohort answers a different question.
Return action
For the rota example, count a manager who publishes another rota during week one. Opening a reminder email, editing the same draft or viewing a colleague’s schedule each measures a different behavior. Several qualifying publications from one manager count as one return.
Observation window
A manager first publishes on Monday, November 2 at 09:00 UTC. Their week one runs from Monday, November 9 at 09:00 UTC up to, and excluding, Monday, November 16 at 09:00 UTC. A return at that exact end time belongs to week two. A manager who starts on Thursday has a Thursday boundary.
Reporting cutoff
Pick the time through which your records are complete. Wait until each person’s entire window ends before adding them to the measured group, including people you have already seen return. Track closed windows with missing evidence separately so an outage cannot silently become a lost user.

Other tools can group users by calendar week or count anyone who returns on or after a particular day. Those choices change which events qualify. Check the time zone, the return rule and the interval before comparing results. Amplitude’s guide to retention windows.

Five of eight shift managers came back

In this fictional example, 12 managers each published their first rota. At the reporting cutoff, three managers still had time left in week one. That left nine completed windows. A logging gap concealed the outcome for one manager. Of the eight with complete records, five published another rota during week one.

We observed 5 returns among 8 managers with known week-one outcomes (62.5%). Three windows are still open and one closed outcome is unknown.

The missing manager could bring the total among all nine closed windows to either 5 of 9 (55.6%) or 6 of 9 (66.7%). That range shows what the missing record can change. When the three open windows close, their outcomes change the cohort total again.

What can you learn from eight people?

With eight observed people, one person moves the rate by 12.5 percentage points. Four returns would give 50%, and six would give 75%. Write down the counts when you share the percentage. A second group that returns at a different rate may have joined through a different channel or had a different opportunity to use the product.

For the rota team, a next step is to ask the three managers with an observed non-return how they staffed the following week. A manager who only writes a monthly rota had a different reason to be absent than one who tried to publish and hit an error. If monthly use is the job, define a monthly window for the next analysis. Keep the original cohort’s definition with its results.

To compare two groups, use the same starting action, return action, elapsed time and observation cutoff rule. A cohort table lets you line up groups at the same age. Mixpanel’s cohort-table walkthrough.

If people have only tried your demo once, the seven-day user test helps you observe their next real opportunity to use it.

Share what you’re building

Send a working title and a short description for private review. Add a deck if you have one.

Open the short application

Applying creates no commitment.