Build in Public

Share SaaS Retention Cohorts While Building in Public

Share SaaS retention cohorts with clear return windows, counts, and unfinished periods. Use a fictional worked example to write an accurate public…

By Uriel Bitton · · 9 min read

Illustrated cover for “Share SaaS Retention Cohorts While Building in Public”.

The short answer

Share retention cohorts by naming the entry rule, counted unit, return action, observation window, and snapshot time. Show counts behind percentages, compare cohorts at the same age, and label unfinished windows. Keep descriptive differences separate from explanations of what caused them.

Share SaaS retention cohorts in public by defining who enters each group, what counts as returning, and how much time each group has had to return. Publish the count behind each percentage and label unfinished observation windows. Compare cohorts at the same age, then explain what the result supports, what remains unknown, and what you will investigate next.

A retention screenshot can make a useful build-in-public update when readers can interpret it without access to your analytics account. The public post should carry the measurement’s meaning with it. A large percentage and a colored cell are not enough to explain whether more customers are finding continuing value.

Start with the retention claim you can actually support

Name the behavior before writing the headline. A workspace publishing another roster, a person opening the app, and a customer renewing a subscription answer different questions. If you measured a return action, describe that action. Reserve claims about paid customer or revenue retention for the corresponding subscription or revenue evidence.

Also define entry into the cohort: the group of people or accounts you follow from a shared starting point. A cohort formed by first signup includes people who never finish setup. A cohort formed by first successful use begins after that step. Both can be useful, but a result from the second group does not describe everyone who signed up.

Choose a consistent unit. If the product serves teams, explain whether you count people or workspaces. Several colleagues can perform the same action inside one account. Your public update should let readers understand whose return you measured and how repeated activity contributes to the count.

These decisions belong in the post’s method note, even when the visual came from a familiar analytics tool. The reader should not have to guess the definition from the tool’s name.

Define the return window beside the result

Specify whether the metric counts activity during a particular interval or activity during that interval and any later time. Those definitions can produce different answers from the same event history.

Amplitude’s retention calculation documentation distinguishes Return On, which counts a return in the specified period, from Return On or After, which also counts later returns. For a public update, retain that distinction in ordinary language. “Published a roster during days 7–13 after first publication” and “published again on day 7 or later” describe different measurements.

Time boundaries also matter. Amplitude’s guide to retention time explains that rolling time windows and calendar dates work differently, and calendar calculations depend on the chart’s time zone. State your own convention rather than assuming “week one” tells the reader enough.

Choose an interval that makes sense for the product’s job. A tool used for weekly scheduling raises a different return question from a tool used for an occasional annual task. The worked example below uses specific weekly windows to make the arithmetic inspectable; it does not establish the right interval for every SaaS.

Keep unfinished periods visible when sharing a chart

A new cohort may not yet have reached a later return window. Its future behavior is unavailable at the snapshot time. Mark that cell with a clear explanation such as “not yet observable,” and give the table an as-of date.

Mixpanel’s retention documentation marks incomplete buckets with an asterisk because their time period is still ongoing and the result remains in flux. When sharing a screenshot or rebuilding a simpler public table, preserve the meaning of those indicators. Cropping out the legend or replacing an unfinished cell with zero changes the story the reader receives.

Check the denominator for the particular interval, too. In an overall retention figure, eligibility can differ by how much time each cohort has had. Amplitude documents that its overall calculations exclude incomplete data. A headline should therefore name the eligible population behind that point rather than attaching the entire signup total to every interval.

The technical documentation cited here was checked on November 15, 2026. Your own export and metric definitions still need checking; a vendor’s description does not verify the data behind your public claim.

A fictional example: two cohorts at different ages

Shift Harbor is an invented SaaS for volunteer coordinators to publish rosters. All workspaces, dates, counts, and public updates in this example are fictional. No customer dataset or product experiment was used for this article.

The founder wants to show whether workspaces return to publish another roster. The example counts each workspace once per return window, regardless of how many members publish or how many rosters they produce. Entry is the workspace’s first published roster, so the table concerns workspaces that reached that milestone.

The calculation uses UTC calendar dates. The first publication date is day 0. The early return window covers days 7–13 inclusive; the later return window covers days 28–34 inclusive. Each row keeps its original cohort size as the denominator for a completed window. A workspace that stops returning stays in that original cohort.

The snapshot is November 15, 2026 at 00:00 UTC. The two cohorts entered on October 1 and November 1 respectively:

First roster published

Original workspaces

Returned during days 7–13

Returned during days 28–34

October 1

8

5 of 8: 62.5%

3 of 8: 37.5%

November 1

10

7 of 10: 70%

Not yet observable

For the October cohort, the early window ran October 8–14 and the later window ran October 29–November 4. Both have ended. For the November cohort, the early window ran November 8–14. Its later window will run November 29–December 5, so that result is unavailable at the snapshot.

An update saying “retention increased from 37.5% to 70%” would compare the older cohort’s later window with the newer cohort’s early window. The figures answer different questions. The November cohort could still return very differently when it reaches the later interval.

The comparable early-window figures are 5 of 8 and 7 of 10, or 62.5% and 70%. That is a descriptive difference of 7.5 percentage points. These small fictional groups do not establish a reliable improvement or explain its cause. One additional returning workspace would move the October percentage by 12.5 points and the November percentage by 10 points.

For the later window, only the October cohort is eligible in this snapshot. Its result is 3 of 8. Dividing those three returns by all eighteen workspaces would incorrectly treat the newer cohort as if its later window had already passed.

Write the public update around the comparable result

Give the reader the finding and the qualification together. A caveat buried below a large improvement headline may not survive when someone quotes the post or shares its image. Keep the cohort sizes, return definition, and unfinished interval close to the result.

Here is proposed public wording for the fictional Shift Harbor example. It is not a real founder report:

At the November 15 snapshot, 7 of 10 workspaces in Shift Harbor’s November 1 cohort published another roster during days 7–13 after their first publication. The October 1 cohort’s equivalent result was 5 of 8.

That gives us a small early-window comparison: 70% versus 62.5%. We cannot yet compare the later window. The October cohort had 3 of 8 workspaces return during days 28–34; the November cohort will not complete that window until December 5 ends in UTC.

We count workspaces, use UTC calendar dates, and count each workspace once per window. These figures describe repeat roster publication among workspaces that already published once. They do not measure every signup or establish paid retention.

The update gives readers a concrete product question to discuss: whether publishing another roster reflects continuing use of the tool’s core workflow. It also creates a useful later follow-up. The founder can return when the newer cohort reaches the same interval, using the same definition.

Separate a cohort difference from a product explanation

A retention table describes recorded behavior under a chosen definition. It does not, by itself, establish why one cohort differs from another. Before attributing a difference to a shipped feature, consider what else changed between the groups.

The cohorts may include different acquisition sources, customer needs, onboarding assistance, or product versions. A change in event tracking can also affect what appears in the table. Record relevant differences privately and include the ones that materially limit the public comparison. Do not invent missing context to make an appealing explanation feel complete.

If the fictional founder introduced a reminder before November, the table alone would not show that the reminder caused the higher early-window figure. A useful public update could describe the reminder as a hypothesis to investigate while keeping the observed cohort comparison separate.

The guide to learning about SaaS churn by building in public covers how public questions can lead to deeper customer investigation. A cohort update can supply a focused question for that work. Replies from other founders may suggest explanations to examine; they do not fill an unfinished window or establish what your customers experienced.

Preserve the method when you return with another update

Keep a private copy of the source export, the query or calculation settings, the snapshot time, and the exact public wording. That makes it easier to understand a later change in the result. Share only the material appropriate for public disclosure.

Before publishing a small cohort, apply the existing guide to deciding which SaaS metrics are safe to share. If counts or group details would reveal information you cannot disclose, use a less detailed account of the learning or defer the numerical update. The fictional counts here illustrate arithmetic, not a recommended minimum size for public reporting.

For the example’s next snapshot, wait until the November cohort’s days 28–34 window has fully ended. State the new snapshot time and report its result against the October cohort’s same window. Keep the original entry and return definitions. If a definition changes, explain that change before treating the new figures as a continuation.

If delayed events or a tracking correction alter an earlier figure, identify the revision and its reason. A changed number can reflect a changed record as well as new behavior. Preserve enough context for readers to understand which occurred.

Before your next retention post, read the proposed headline beside the cohort table. Check that it refers to the same population, action, and elapsed interval. Then publish the useful result with its unfinished parts still visible.

Questions founders ask next

Does a completed window mean the data can never change?

No. The observation period can end while delayed events, tracking repairs, or data corrections still change the recorded result. Keep the snapshot date and describe material revisions. Completion means the specified time has elapsed, not that every data-quality question is settled.

What if the SaaS has only one mature cohort?

Report that cohort descriptively if the data is suitable for disclosure. State its entry rule, return action, size, window, and snapshot date. You can explain what you will compare later without claiming a trend before comparable cohorts exist.

Can users return after missing a particular window?

Yes. A workspace that misses a defined interval can perform the action later. Report what your chosen measure counts. An interval-specific return result does not by itself mean the workspace permanently stopped using the product.

Sources

Frequently asked questions

Does a completed window mean the data can never change?

No. The observation period can end while delayed events, tracking repairs, or data corrections still change the recorded result. Keep the snapshot date and describe material revisions. Completion means the specified time has elapsed, not that every data-quality question is settled.

What if the SaaS has only one mature cohort?

Report that cohort descriptively if the data is suitable for disclosure. State its entry rule, return action, size, window, and snapshot date. You can explain what you will compare later without claiming a trend before comparable cohorts exist.

Can users return after missing a particular window?

Yes. A workspace that misses a defined interval can perform the action later. Report what your chosen measure counts. An interval-specific return result does not by itself mean the workspace permanently stopped using the product.

A note from Uriel Bitton

I share practical lessons from building Buildside and learning alongside SaaS founders.

Keep building with us

Get practical founder lessons and product updates in your inbox.

Subscribe or Join Buildside.

Keep reading

Browse more articles from Buildside