Read customer retention by cohort

Choose a starting point that matches the question

A cohort is a group joined by a specified event, not simply a convenient list in the current customer system. For paid retention, you might begin at the first completed purchase. For product learning, you might begin at the first useful task. Decide which question matters and name the event clearly. Mixing trial registrations with paying customers creates a denominator whose meaning shifts between acquisition campaigns.

Select a retention event suited to the expected use pattern. A weekly planning service may reasonably examine weekly plan activity, while an annual compliance review cannot be judged by daily logins. Specify whether the measure requires an active subscription, a completed task or both. Document cancellations, pauses and reactivations. Keep organisation-level and person-level measures separate when several employees use one purchased account; otherwise staff turnover can resemble customer loss.

Compare customers at the same age

Build rows by entry period and columns by elapsed age, such as the end of each completed month after entry. Count eligible customers at each age against the original cohort denominator. Do not shrink the denominator by removing people who cancelled. Show unavailable cells for windows that have not finished rather than filling them with zero or assuming all recent customers are retained.

Write the cutoff date beside the table and make the interval rule reproducible. “Month two” might mean a calendar month after signup or the second thirty-day period; either can be useful if consistently defined. Check a few records manually, including a cancellation followed by a return. For a small cohort, show counts prominently because a single customer may shift the percentage substantially. Avoid ranking segments whose definitions or observation periods differ.

Use the table to ask a better question

When two rows differ, examine acquisition route, offer version, seasonality and onboarding assistance before making a causal claim. A larger cohort can include more people with occasional needs. It can therefore contain more active customers while retaining a smaller proportion. Segment only where the records and sample size support a meaningful comparison; repeated slicing can produce dramatic but fragile stories.

Pair the quantitative review with a few specific journey checks. Look for changes in the task customers wanted to complete, support events or predictable project endings. Record a hypothesis and a bounded investigation rather than immediately rebuilding the service. SBA’s marketing material connects customer experience with repeat business; cohort analysis is one practical method for examining that relationship. It remains descriptive evidence, not proof that a particular feature caused a return.

Keep the original data extract and calculation notes with the table. If the retention definition changes, rerun earlier rows under the new rule or mark a break in the series. A rising line is difficult to interpret when the team quietly switches from completed work to any login midway through reporting.

TAKE THIS INTO YOUR NEXT CONVERSATION

A practical next-step checklist.

• Define entry and meaningful retained activity.
• Compare equal customer ages with a fixed denominator.
• Mark incomplete periods as unavailable.
• Review customer mix before proposing a causal explanation.