Segmentation answers: do different customers behave differently.
Example: customers who adopted feature A have two percent churn. Customers who did not adopt feature A have four percent churn. Feature A drives retention. Feature A is important.
Cohort analysis answers: do different groups of customers behave differently over time.
Example: customers who signed in January have eighty percent retention in month one, sixty percent in month two, forty percent in month three. Customers who signed in February have seventy-five percent retention in month one, fifty percent in month two. January cohort retains better. Something changed between January and February.
When to use segmentation
Use segmentation when metric varies by segment
Revenue varies by customer size (enterprise vs self-serve). Traffic varies by traffic source (paid vs organic). Engagement varies by product usage (power users vs occasional users). Segment and measure separately. Understanding variance improves decisions.
Do not segment if metric is same across segments
Page views are same on Monday and Tuesday. No point segmenting by day. Time is not a useful segment.
When to use cohort analysis
Use cohort analysis when cohort matters
How does customer lifetime value differ by sign-up month. Months matter (seasonality, product changes). Cohort analysis reveals time effects.
Do not use cohort analysis if cohort is meaningless
Pages loaded per visit does not vary meaningfully by sign-up month. Cohort analysis would show noise, not signal.
Real example: advanced segmentation at SaaS company
Initial segmentation by size
Enterprise (annual contract value one hundred thousand plus): twenty customers. Two percent monthly churn. Mid-market (ten thousand to one hundred thousand): one hundred customers. Four percent monthly churn. Self-serve (under ten thousand): five hundred customers. Eight percent monthly churn.
Enterprise customers are sticky. Self-serve customers churn fast. This is expected. But mid-market churn at four percent is above target of three percent.
Further segmentation by behavior
Further segment mid-market by product adoption. Mid-market customers using feature A (adoption): two percent monthly churn. Mid-market customers not using feature A (no adoption): six percent monthly churn.
Feature A is retention driver for mid-market. Action: improve feature A discovery for mid-market customers. Get more mid-market customers to adopt feature A. Churn should decrease.
Real example: cohort analysis at marketplace
Track buyer repeat rate by sign-up cohort
January cohort (one thousand buyers): repeat rate month one is thirty percent. Month two is twenty percent. Month three is fifteen percent.
February cohort (one thousand buyers): repeat rate month one is twenty-five percent. Month two is eighteen percent. Month three is twelve percent.
Discover difference
January cohort retains better than February. Why. Investigate.
Test hypothesis
Hypothesis: January had special onboarding. February did not. Test: reintroduce onboarding for March cohort.
March cohort (one thousand buyers): repeat rate month one is thirty percent (similar to January). Month two is twenty-two percent. Month three is sixteen percent.
Onboarding improved retention. Feature A was the fix.
Segmentation and cohort analysis together
Combine both for powerful analysis
Segment enterprise customers by cohort.
January enterprise cohort: eighty-five percent renewal rate. February enterprise cohort: eighty percent renewal rate. March enterprise cohort: ninety percent renewal rate.
Investigate and act
March enterprise is renewing better. Why. Investigate. Find: March cohort was onboarded with new process that emphasizes relationship building. Relationship building increases enterprise renewal rates. Action: apply new onboarding to all cohorts going forward.
Frequently asked questions
How many segments should we track?
Three to seven segments maximum. More segments create complexity. Sweet spot is three to five. Example: customer size (enterprise, mid-market, self-serve) is three segments. That is clean. Adding geography (US, EU, APAC) creates nine segments (3x3). Too many. Keep segments to one or two dimensions. With WEMASY's website builder, you can set this up directly on your website.
How do we decide which dimension to segment by?
Segment by dimension that affects behavior. Customer size affects behavior (enterprise behaves differently from self-serve). Industry affects behavior (SaaS customers behave differently from retail). Traffic source affects behavior (paid search behaves differently from organic). Segment by dimension that explains variance in metric.
How far back should cohort analysis go?
At least twelve months. Shorter cohorts (less than three months old) are too immature. Longer cohorts (twelve months plus) show full lifecycle. Example: track customer churn out to twelve months because annual contracts are commitment. Track buyer repeat rate out to six months because buying cycle is shorter.
Should we use dollar amounts or percentages in cohort analysis?
Percentages for retention/repeat metrics. Dollar amounts for revenue/value metrics. Retention is same whether customer is small or large (both need to remain as percentage of cohort). Revenue varies by customer size (need dollar amounts to see impact).
How do we handle seasonality in cohort analysis?
Account for it. January cohort has holiday carryover. April cohort is post-tax season. Cohorts differ due to seasonality, not product changes. Compare same-season cohorts. January 2024 cohort compared to January 2025 cohort. Year-over-year removes seasonality.
How do we communicate cohort insights?
Show cohort retention table with columns for month one, month two, month three. Show rows for each cohort. Pattern jumps out. January cohort has eighty, sixty, forty (steady decline). February cohort has seventy, fifty, thirty (steeper decline). Visual table is clearer than written explanation.