Cohort Analysis for Ecommerce Retention

Why a single aggregate retention number misleads

"Our repeat purchase rate is 24%" sounds like a fixed fact about the business, but it's actually a blend of every customer who's ever purchased, regardless of when. A business that's rapidly acquiring new customers will show a lower blended repeat-purchase rate even if retention behavior is improving, simply because a growing share of the customer base is too new to have had a realistic chance to repeat-purchase yet. Cohort analysis fixes this by grouping customers by when they were acquired and tracking each group's behavior separately over time, so you're comparing like-aged groups to each other rather than blending customers of wildly different tenure into one number.

What a cohort is, concretely

A cohort is simply a group of customers who share an acquisition period — commonly a month ("the March cohort" = everyone who made their first purchase in March). You then track that specific group's behavior going forward: what percentage made a second purchase within 30/60/90 days, what percentage are still active after six months, what their average cumulative spend looks like at each milestone.

Building a basic cohort table

Reading down a column (comparing the same milestone across cohorts) tells you whether retention behavior is improving, stable, or declining over time — the March cohort's 11% one-month repurchase rate compares directly to January's 8% and February's 9%, showing a genuine upward trend rather than noise. Reading across a row tells you how a single cohort's behavior develops as it ages.

What drives a cohort's retention trend, worth investigating

Acquisition channel/campaign changes: a shift toward a channel that brings lower-intent, deal-driven customers can lower repeat-purchase rates for cohorts acquired during that period, even if overall business metrics (revenue, order volume) look fine on the surface. Product or category mix at acquisition: customers whose first purchase was a genuinely repeat-use consumable will structurally show higher repurchase rates than customers whose first purchase was a one-time durable good — comparing cohorts without accounting for this can create a misleading impression that "retention is declining" when it's actually a product-mix shift. Post-purchase experience changes: a change in email/lifecycle marketing, loyalty program launch, or a customer service quality shift can show up as a visible inflection point in a specific cohort's trend line.

A worked example: catching a problem early with cohorts

A seller notices their blended repeat-purchase rate has been flat at roughly 22% for the past year, so retention looks stable at a glance. Breaking it into monthly cohorts reveals that each new cohort's one-month repurchase rate has actually been declining steadily (from around 12% a year ago to around 7% recently), while older cohorts continue accumulating repeat purchases over a longer window and propping up the blended average. The blended number hid a real, ongoing decline in how well new customers are being retained — something worth investigating in acquisition channel mix or early post-purchase experience — that wouldn't have been visible without the cohort breakdown.

Where this connects to marketplace selling specifically

Cohort analysis is most natural on a direct-to-consumer channel where you have durable customer identity across purchases. On most marketplaces, you have limited-to-no visibility into individual repeat customers by design (the marketplace, not the seller, generally owns the customer relationship) — so this technique applies most directly to your own website's customer base, or to marketplace programs that specifically expose some form of repeat-customer data. If most of your revenue is marketplace-based, a rough proxy is tracking repeat-purchase-rate-equivalent signals the marketplace does expose (subscribe-and-save enrollment, brand-follow counts) rather than true cohort-level tracking.

Best practices

• Compare cohorts at the same milestone (month 1 to month 1, month 3 to month 3) rather than comparing a mature cohort's total lifetime number to a brand-new cohort's early number.
• Segment cohorts by acquisition channel when you have enough volume, since channel quality is one of the most common real drivers of a retention trend.
• Revisit product/category mix at acquisition before concluding a retention trend reflects a genuine customer-experience problem.
• Apply this primarily to your owned/direct channel, where you have real customer-level visibility, rather than trying to force it onto marketplace channels that don't expose the underlying data.

Checklist

• Define cohort grouping (typically by acquisition month)
• Choose milestones to track (e.g., 30/60/90/180 days)
• Build the cohort table and compare same-milestone values across cohorts
• Segment by acquisition channel if volume allows
• Investigate any cohort showing a meaningfully different trend from its neighbors

FAQs

How much customer data do I need before cohort analysis is worthwhile? You need enough repeat-purchase history to have at least a few cohorts reach your chosen milestones — a business only a few months old, or one with very low repeat-purchase volume, won't yet have enough signal for this to be more informative than the aggregate number.

Should I use weekly, monthly, or quarterly cohorts? Monthly is the most common default — weekly cohorts are often too small to show a stable pattern unless volume is high, and quarterly cohorts can smooth over a meaningful within-quarter shift.

Does cohort analysis apply to advertising spend as well as retention? Yes — the same acquisition-period grouping logic is commonly used to track cumulative revenue or margin per cohort against acquisition cost, which is closely related to (and often analyzed alongside) Attribution and Incrementality Across Owned + Marketplace Channels.