Aug 14, 2026

Cohorts Thresholds Worth Writing Down

A cohort is a group of customers or orders segmented by a shared attribute (acquisition date, traffic source, product purchased) tracked across time periods to measure retention, repeat purchase rate, or lifetime value. Cohort analysis isolates the effect of timing or source on customer behavior by holding the cohort constant while measuring outcomes across subsequent periods.

Why Cohorts Matter for DTC Operations

Aggregate metrics (overall repeat purchase rate, average order value) hide critical patterns. A 25% repeat purchase rate tells an operator nothing about whether Q1 customers are better or worse than Q4 customers, or whether email-sourced customers outperform paid search. Cohort analysis isolates these differences by grouping customers by acquisition date or source, then tracking their behavior in subsequent periods.

For Shopify teams, cohorts answer three operational questions: (1) Are newer customers worse? (2) Did a traffic source or campaign change customer quality? (3) Is a retention or repeat purchase problem recent or structural? Without cohorts, these questions require guesswork.

Cohort Definition Checklist

Before building a cohort table, lock down the cohort attribute and the metric being tracked.

Cohort attribute: acquisition date (weekly, monthly), traffic source, product category, email list segment, or paid campaign. Single attribute per table.
Time period: days, weeks, or months post-acquisition. Monthly is standard for DTC; weekly for high-frequency repeat purchases (beauty, supplements).
Metric: repeat purchase rate (% of cohort that ordered again in period N), repeat purchase count (average orders per customer), or LTV (revenue per customer). Pick one metric per table.
Denominator: new customers only (exclude existing customers), or all orders (include repeat). State explicitly.
Lookback window: minimum 3 periods (e.g., 3 months) to detect trend. 12+ periods preferred for seasonal validation.

Failure Mode: Cohort Size and Noise

Small cohorts produce unreliable thresholds. A cohort of 50 customers with 2 repeat purchases (4% repeat rate) is not comparable to a cohort of 5,000 customers with 200 repeat purchases (4% repeat rate). The first is noise; the second is signal.

Threshold: Reject cohorts with fewer than 100 customers in the first period. For paid traffic sources (Facebook, Google), require 200+ customers per cohort to reduce variance. For organic or email, 100 is acceptable if the trend is consistent across 3+ periods.

Failure Mode: Survivorship Bias and Churn

Cohort tables often hide churn. If a cohort acquired 1,000 customers in Month 1, but only 300 are still active (not churned) by Month 3, the repeat purchase rate in Month 3 is calculated on 300, not 1,000. This inflates the repeat rate because low-engagement customers have already left.

Decision rule: Always include a churn column (% of cohort still active) before the repeat purchase rate column. If churn exceeds 70% by period 3, the repeat purchase rate is unreliable. Flag the cohort as 'high churn - metric suspect.'

Threshold: Cohort Decay and Acceptable Decline

Repeat purchase rates decline over time. A cohort acquired in January may have 40% repeat purchase rate in February (period 1), 25% in March (period 2), and 15% in April (period 3). This is normal. The question is: how fast should decay occur?

Benchmark thresholds for DTC Shopify brands:

Period 1 repeat rate (first 30 days): 15% - 35% is healthy. Below 10% signals poor product-market fit or onboarding failure. Above 40% suggests high-frequency replenishment (supplements, beauty) or subscription model.

Period 2 repeat rate (30 - 60 days): 60% - 75% of period 1 rate. If period 1 is 25%, expect 15% - 19% in period 2.

Period 3 repeat rate (60 - 90 days): 40% - 60% of period 1 rate. If period 1 is 25%, expect 10% - 15% in period 3.

If decay is steeper (e.g., period 2 is 40% of period 1), investigate product quality, email cadence, or fulfillment issues.

Threshold: Cohort Comparison and Statistical Significance

Comparing two cohorts (e.g., 'paid search' vs. 'organic') requires a decision rule. A 2 - 3 percentage point difference in repeat purchase rate is often noise, not signal.

Threshold: Require a minimum 5 percentage point difference in period 1 repeat rate to flag a cohort as 'different.' For example, if organic cohorts average 28% repeat rate and paid search averages 22%, the 6 - point gap is worth investigating. A 2 - point gap is not.

For LTV comparisons, require a minimum 15% difference (e.g., organic LTV of $120 vs. paid search LTV of $100 is noise; $120 vs. $85 is signal).

Failure Mode: Seasonal Cohorts and Year - Over - Year Drift

A cohort acquired in November (holiday season) will have different repeat purchase behavior than a cohort acquired in June. Comparing them directly is invalid. Similarly, a cohort from November 2023 may not be comparable to November 2024 if customer acquisition cost, product mix, or email strategy changed.

Decision rule: Compare cohorts acquired in the same calendar month or season across years. If comparing across seasons, note the seasonal adjustment. If a cohort's repeat purchase rate is materially different from the prior year (>10 percentage points), investigate changes to acquisition, product, or retention strategy.

Questions

FAQ

When should a DTC operator stop using cohort analysis?

When cohort size drops below 100 customers per period, or when churn exceeds 80% by period 2. At that point, the cohort is too small or too degraded to produce reliable thresholds. Switch to individual customer tracking or segment analysis instead.

What's the difference between cohort analysis and segment analysis?

Cohort analysis groups customers by acquisition date or source and tracks them forward in time. Segment analysis groups customers by a static attribute (e.g., 'high-value' vs. 'low-value') and compares behavior at a single point in time. Cohorts answer 'how do customers change over time?' Segments answer 'how do different customer types behave now?'

Should repeat purchase rate be calculated on all customers or only those who made a first purchase?

Only customers who made a first purchase. The denominator is the cohort size (e.g., 1,000 customers acquired in January). The numerator is the count who purchased again in period N. This avoids inflating the rate by excluding non-purchasers.

How often should cohort tables be updated?

Monthly for most DTC brands. Weekly updates are useful only if repeat purchase cycles are very short (e.g., subscription or daily-use products). Quarterly reviews are sufficient for decision-making if the business is stable. Update whenever a major change occurs (new product launch, email strategy shift, traffic source change).

Want this on your account?

Thirty minutes. Bring the number that keeps you up.

Book a demo