Design a behavioral cohort diagnosis

Aggregate Retention Hides Everything

Aggregate retention curves lump cohorts with radically different profiles, producing meaningless averages. Amplitude's cohort analysis writing and Mixpanel's retention guidance both argue that cohort segmentation is non-optional — the top 20% and bottom 20% cohorts usually behave so differently that aggregate metrics conceal both the success story and the leak.

How the Design a behavioral cohort diagnosis Prompt Works

The prompt slices cohorts on four dimensions (acquisition channel, first-week behavior, segment, signup month), produces retention curves per cohort, and identifies the cohorts that drive aggregate numbers. The "what if we only kept the top 3 cohorts" question clarifies which cohorts are actually worth acquiring.

When to Use It

• Aggregate retention is flat or declining and the cause is unclear.
• Paid acquisition is rising and retention is not.
• A feature launch seems to help some users and hurt others.
• A new PM wants a baseline for activation and retention interventions.
• Leadership is asking "where is the growth leak?"

Common Pitfalls

• Looking only at aggregate retention. Aggregate hides the winners and the leak. Always segment.
• Cohort count >5. Too many cohorts and every slice looks like noise. Start with 3-4 high-signal dimensions.
• Ignoring seasonality. Signup month matters. Cohort comparisons without time controls are unreliable.

Sources

• Amplitude Blog — Amplitude
• Mixpanel Blog — Mixpanel
• Retention, Engagement & Growth: The Silent Killer — Reforge
• The Activation Metric — Reforge