Write a narrative summary of a cohort analysis
intermediateClaude SonnetData & AnalyticsData Analystcohort-analysisretentionproduct-analyticsnarrativedata-analyst
Use case
Use this prompt when you have a cohort retention table (CSV, SQL output, or pasted grid) and need to deliver a written analysis to product, growth, or leadership. A retention grid alone rarely changes minds — the narrative on top is what drives action.
The prompt
You are a senior data analyst writing the narrative summary of a cohort retention analysis. Your reader is a product or growth lead who can read numbers but does not have time to stare at the grid themselves. Inputs: - Cohort table:{{cohort_table}}(rows are cohorts by signup month/week, columns are periods since signup, cells are retention %) - Cohort definition:{{cohort_definition}}(e.g., "all users who completed signup", "paid trial starters", "activated accounts") - Retention metric:{{retention_metric}}(e.g., "weekly active", "made a payment", "logged in") - Time range:{{time_range}}- Known product or marketing changes during the window:{{product_changes}}- Audience for the writeup:{{audience}}Write a narrative with these sections: ## Headline (2–3 sentences) The single most important pattern in the data, stated directly. If retention is improving, by how much and starting when. If it's flat or worsening, say so. ## Cohort Behavior - Compare early-period retention (period 1–2) across cohorts — is the top of the curve moving? - Compare late-period retention (period 6+ or whatever is available) — is the long-tail floor changing? - Identify any cohorts that broke from the pattern, positively or negatively, and propose a reason tied to{{product_changes}}when plausible. ## What's Likely Driving It Two to four hypotheses for the trend, ranked by how well they fit the data. Be explicit about which evidence supports each — and which would falsify it. Do not stretch causation beyond what the cohort grid alone can support. ## What This Means Translate the pattern into business consequence. If retention is up X points at week 4, what does that imply for LTV, payback, or the next planning cycle? Use rough math, not false precision. ## Recommended Next Cuts Suggest 2–3 follow-up cuts that would sharpen the analysis — by acquisition channel, plan tier, persona, or onboarding path. Be specific about which cut tests which hypothesis. Tone: direct, confident where the data is strong, hedged where it isn't. Do not use the word "interesting." Do not pad. If a section has nothing meaningful to say, write one sentence and move on.
Variables
{{{{cohort_table}}}}Replace with your {{cohort table}}{{{{cohort_definition}}}}Replace with your {{cohort definition}}{{{{retention_metric}}}}Replace with your {{retention metric}}{{{{time_range}}}}Replace with your {{time range}}{{{{product_changes}}}}Replace with your {{product changes}}{{{{audience}}}}Replace with your {{audience}}
Example
Input
cohort_table: | Cohort | M0 | M1 | M2 | M3 | M4 | M5 2024-01 | 100% | 42% | 31% | 26% | 24% | 23% 2024-02 | 100% | 41% | 30% | 25% | 23% | 22% 2024-03 | 100% | 44% | 33% | 28% | 26% | 25% 2024-04 | 100% | 48% | 37% | 31% | 29% | - 2024-05 | 100% | 52% | 40% | 33% | - | - 2024-06 | 100% | 53% | 41% | - | - | - cohort_definition: New signups who completed email verification retention_metric: Logged in at least once during the calendar month time_range: January–June 2024 cohorts product_changes: New onboarding checklist shipped Apr 3; in-app tour redesign May 12; pricing page change May 20 (no plan changes) audience: VP of Product and the growth PM
Output
## Headline M1 retention jumped from a 41–44% range in Q1 to 52–53% for the May and June cohorts — roughly a 10-point lift that holds through M2 and M3. The improvement appears to start with the April cohort and accelerates after the May 12 in-app tour redesign. ## Cohort Behavior Q1 cohorts (Jan–Mar) cluster tightly: M1 retention 41–44%, M3 retention 25–28%. The April cohort breaks the pattern at M1 (48%), and May and June extend the lift to 52% and 53%. Late-period retention is also rising — M3 for the April cohort is 31% versus 25–28% for Q1 cohorts, suggesting the gain is not just an activation bump that decays. Too early to call the M4 floor, but the April M4 of 29% is meaningfully above Q1's 23–24% floor. No cohort moves backwards. The lift is monotonic from March onward, which makes a one-time external cause less likely. ## What's Likely Driving It 1. **In-app tour redesign (May 12) — strongest fit.** May and June cohorts each saw 4-point M1 jumps over the prior cohort, lining up with the tour ship date. Falsifier: if April-cohort users who never saw the tour also retained at 48%, the tour is not the driver. 2. **Onboarding checklist (April 3) — partial fit.** April's 4-point M1 lift over March aligns with the ship date. The checklist plausibly improved activation in a way that compounds with the tour. 3. **Acquisition mix shift — possible but unverified.** A change in channel mix (e.g., more product-led signups, fewer paid social) could explain durable retention gains without any product change. This needs a channel cut to rule in or out. 4. **Seasonality — unlikely.** Spring/summer seasonality in this product has historically been small (<2 points), and the lift exceeds that. ## What This Means If the new M3 retention of ~31% holds through M6 at the same Q1 decay rate, the long-run retention floor for May–June cohorts lands near 27–28%, versus 22–23% for Q1. On a rough LTV basis, that's a 20–25% lift in expected lifetime sessions per signup — material enough to revisit payback assumptions in the FY plan if it sustains for two more cohorts. ## Recommended Next Cuts - **Acquisition channel cut on May–June cohorts.** Tests the channel-mix hypothesis directly. - **Tour-exposed vs. tour-unexposed users within the May cohort.** Cleanest test of the tour hypothesis. Use the natural A/B from the staged rollout if one exists. - **Plan tier cut on April–June cohorts.** Tells you whether the lift is concentrated in self-serve or also showing up in trial-to-paid behavior.
Tips for best results
- 1Always lead with the headline number. Readers will skim — make the first paragraph carry the analysis.
- 2Check that cohorts have comparable size before declaring a trend. A 50% retention rate on 80 users is not the same evidence as 50% on 8,000.
- 3When you have a candidate cause and a candidate ship date, state explicitly what evidence would falsify the link. Analysts gain credibility by ruling things out, not just in.
- 4Late-period cohort cells are partial — be careful not to compare a partial M4 to a complete M4. Mark unfinished periods as incomplete.
- 5If you only have 3–4 cohorts post-change, hedge. Cohort patterns can reverse with 2 more cohorts of data.
Related prompts
Write up a retention analysis with cohort-level insight
advancedProduce a structured retention analysis writeup that goes beyond the cohort table — segments by behavior, isolates drivers, and recommends concrete retention bets.
Data & Analyticsretentioncohort-analysisproduct-analytics
Generate weekly metrics commentary from a CSV
beginnerTurn a weekly metrics CSV into a tight written commentary that explains what moved, why, and what to do about it.
Data & Analyticsweekly-metricsbusiness-reviewcommentary
Generate exec narrative on top of a dashboard
advancedProduce a one-page exec narrative that sits on top of a dashboard data dump — the story leadership reads instead of squinting at the dashboard.
Data & Analyticsexecutive-reportingkpinarrative
Need help implementing this prompt in your workflow?
Book a call