Cohort Analysis

Cohort analysis groups customers by a shared starting characteristic — almost always their first-purchase period — and tracks how each group behaves over time: whether they return, how much they spend, and when they stop. In ecommerce it is the measurement engine beneath Retention, Churn Rate, Customer Lifetime Value (CLV) and Subscription Commerce: those metrics are only trustworthy when read by cohort rather than in aggregate. This page was harvested web-only (run 116) — both practitioner streams were down — so every benchmark below traces to an agency- or vendor-authored source and should be read as such.

Firewall: every claim is what a source reports. See ../../CONTEXT.md Rule 1.

What it is and why it beats aggregate metrics

Niblin describes a cohort as a group of customers who first purchased in the same time period, and cohort analysis as tracking how each group behaves over time — whether they return, how much they spend, and when they stop (Niblin, niblin.com, as-of 2026-05-25). The same source argues that headline LTV hides the story: an LTV of $200 doesn't reveal whether you have 100 customers buying twice or 10 buying 20 times — "cohort analysis reveals the shape of your retention, which LTV hides" (Niblin, 2026-05-25).

Retention-Led Growth distinguishes two kinds of retention — aggregate (Shopify's "Returning Customer Rate") and cohort — and reports that using the aggregate figure alone produces survivor bias: the metric is dominated by older loyal customers while masking churn among newer cohorts (Retention-Led Growth / Marcin Mleczko, as-of 2023-12-18). It gives a worked illustration: aggregate retention can appear to climb from 45% to 70% over five years not because each customer became more likely to stay, but because naturally loyal "good" customers come to dominate the base — the fix being to track retention by first-purchase cohort.

Userpilot adds that aggregate churn is distorted by early "tourist" churn that makes retention look worse than it really is; rebasing to a later period (e.g. Month 3) gives a cleaner read of real users, and cohort analysis shows whether a retention problem is universal or concentrated in specific segments (Userpilot, userpilot.com, as-of 2026).

How to read a cohort table

Niblin's reading guide (as-of 2026-05-25): rows = cohorts (by first-purchase month), columns = months after first purchase ("customer age", Month 0 = the purchase month), and cells = the % of that cohort who purchased again. Read left-to-right to see decay over time within one cohort; read top-to-bottom to see whether newer cohorts retain better or worse than older ones.

A typical ecommerce retention curve, per Niblin and Userpilot, starts at 100% (everyone made a first purchase), drops steeply in the first 30–60 days as one-time buyers reveal themselves, and stabilises at roughly 15–25% of the original cohort after 12 months (as-of 2026; benchmark, med confidence).

Curve shapes practitioners name

Userpilot's taxonomy (as-of 2026, single-source / med confidence): "flatten-and-hold" (drops 4–8 weeks then plateaus on a foundational core), "smile curve" (dips, flattens, then rises as lapsed users return), and "perpetual decline" (steep drop that keeps falling past month 3 with no stabilisation).

Niblin defines four cohort patterns with diagnoses (as-of 2026-05-25):

Pattern	Niblin's diagnosis
Steep drop then flat	Most never return but a loyal core forms → fix the second purchase
Gradual perpetual decline	Weak product-market fit or no retention mechanism → fix product / add loyalty
Cohorts declining top-to-bottom	Customer quality degrading (scaling ads into broader, worse-fit audiences)
Cohorts improving top-to-bottom	Product/targeting improving → scale what works

Benchmarks (as-of 2026-02-14 unless noted)

All figures below are vendor/agency-sourced and volatile. BS&Co's are from a single agency portfolio of 156,110 DTC customers; treat ranges, not point values, as the signal.

• Aggregate repeat-purchase rate ≈ 18.8% within a 365-day window across the 156k-customer sample — i.e. 81.2% never buy a second time; brand range 7.1%–44.2% (BS&Co, bsandco.us, as-of 2026-02-14).
• By category (BS&Co): consumables 22–44% (typical 30–40%); fashion/apparel/beauty/jewelry 10–17% (apparel clusters 15–17%, high-AOV jewelry ~11%); durables/general retail 7–18%.
• Niblin's "rough" D2C cohort guide (self-described, med confidence, as-of 2026-05-25): Month 1 repeat <8% below average / 8–15% average / >15% strong; Month 3 cumulative <15% / 15–25% / >25%; Month 6 <20% / 20–35% / >35%; Month 12 <25% / 25–40% / >40% (heavily category-dependent).

Time-to-second-purchase (drives post-purchase timing)

Of customers who do repurchase, BS&Co reports 50.3% do so within 30 days and 76.4% within 90 days; only 3.7% take longer than a year (same-day 6.3%, within 1 week 15.9% cumulative) (bsandco.us, as-of 2026-02-14). By category, fashion/apparel has the fastest median time to second purchase at 15–27 days (seasonal needs + gifting) vs consumables 27–68 days.

BS&Co's methodological point: median time-to-second-purchase (15–35 days) and the average (50–100+ days) diverge sharply because a long tail of late returners inflates the average — "the median is the truth"; planning post-purchase strategy around the average means being too late for most potential repeat buyers.

Revenue cohorts vs retention cohorts

Niblin notes that Shopify's native cohort tool shows repeat-purchase rate (% who bought again) but not revenue-based cohorts — a 12% Month-1 retention is meaningless if those buyers spent $5 in repeat revenue against a $200 first order; revenue cohorts require third-party tools (niblin.com, as-of 2026-05-25). Triple Whale's cohort charts group customers by first-order date and show revenue generated over time (a revenue-cohort view), sliceable by acquisition channel/creative.

Cohorts can also be sliced beyond first-purchase month. Retention-Led Growth gives a DTC example where slicing by subscription plan and acquisition source showed Premium-plan subscribers declined slower (higher M12 retention) while Flash-Sale-acquired customers churned markedly faster — i.e. discounted-acquisition cohorts retain worse (retentionledgrowth.com, as-of 2023-12-18; illustrative, not a benchmark).

Common mistakes (as reported)

• Survivorship bias — analysing only surviving/active cohorts overlooks churned customers; the mitigation cited is to include both active and inactive cohorts and track everyone forward (intent-to-treat) so the full denominator is preserved (Adasight / Retention-Led Growth).
• Suppressing recent buyers for 30–60 days ("don't annoy them") mutes marketing during exactly the window when 50.3% of repeat purchases happen — the data contradicts the instinct (BS&Co, 2026-02-14).
• Optimising post-purchase flows for cross-sell when BS&Co finds 77% of second purchases are reorders of the same product and only 23% are cross-sells (apparel reorder 48–66%; supplements 82–93%; home decor ~0% reorder). (BS&Co flags apparel reorder data may include exchanges processed as new orders.)
• Too-short email flows — most post-purchase flows run 7–14 days, covering only ~16–30% of the repeat-purchase window; BS&Co argues for extending to 90 days to cover 76.4% of repeat purchases (vendor advice framing, recorded as their claim).

Tooling (as-of 2026-05-25; pricing volatile, vendor COI)

Niblin reports native Shopify cohort analysis is Shopify Plus only; on Growth/Pro plans teams use Lifetimely (~$49/mo), Peel Insights ($200+/mo), Triple Whale (~$199/mo, scales with revenue), or Niblin itself. Niblin lists Shopify-native blind spots: no cohort-by-acquisition- source, no revenue-based view (only repeat rate), no predictive LTV modelling, no churn alerting, and basic plain-table visualisation (no decay curves/heatmaps). Putler/Zigpoll position Lifetimely on LTV/profit/acquisition-source cohorts and Triple Whale as broader DTC analytics/attribution with revenue cohort charts (vendor-review sources, med confidence).

Supporting economics (historical context)

Key terms

Term	Meaning (as reported)
Cohort	Customers sharing a start characteristic, usually first-purchase month
Customer age	Months elapsed since first purchase (the column axis; Month 0 = purchase month)
Aggregate vs cohort retention	Whole-base returning-customer rate vs first-purchase-cohort tracking; aggregate masks newer-cohort churn (Retention-Led Growth)
Survivor bias	Distortion from measuring only retained/active customers (Retention-Led Growth)
Revenue cohort	Cohort tracked by repeat revenue over time, not just repeat rate (Niblin)
Keep rate	Share of a cohort still active in a later period (the survival-curve value)

Gaps / open questions

• No first-party or academic source fetched — material is agency/vendor (BS&Co, Niblin, Userpilot, Envive) carrying conflict-of-interest; McKinsey/Bain/Amplitude/Reforge/Lenny's not reached directly. A Bayesian cohort-modelling paper (arXiv 2504.16216, "Cohort Revenue & Retention Analysis: A Bayesian Approach", 2025) was logged but not fetched — a lead for a future pass.
• No fashion-specific named case study (e.g. a UNIQLO/Zara/DTC-apparel cohort breakdown); benchmarks are category-aggregate.
• Both practitioner streams down — reddit-research MCP and Apify transcript actor not connected, so no operator counter-narrative on which cohort window/horizon brands actually use. Five higher-signal YouTube candidates were logged for a transcript-enabled re-run (lead DnpYApVNu-Y Shopify cohort guide; cQ18pgPy-uE "Cohort Retention is NOT the Only Retention Method").
• UK/Europe-specific cohort/repeat-purchase benchmark unfilled — a genuine data gap for UNIQLO Europe (consistent with the run-115 retention gap).