Cohort Analysis for Retention Marketers: How to Read Repeat Purchase Curves and Spot Where Customers Drop Off

Last updated: October 2026

Ecommerce cohort analysis for retention only works when you stop slicing by signup month alone. Split cohorts by acquisition channel, first product, and first-order offer, read the shape of each repeat purchase curve, and match where it drops to the flow responsible. That turns a declining line into a fix list.

Most brands build a cohort chart once, see a line sloping down, conclude "retention is bad," and go back to planning campaigns. Nothing changes, because nothing in the chart told them what to change.

A cohort analysis is an X-ray, not a report card. The diagnosis lives in two places most teams never look: how you cut the cohort and the shape of the curve. Get both right and the chart points at a specific flow, offer, or product.

What does a cohort analysis actually tell a retention marketer?

A cohort analysis tells you when customers stop coming back, not just how many do. By grouping customers who share a starting point and tracking what share places another order over time, you see whether retention breaks early, late, or gradually, which is what you need to pick the flow to fix.

Cohort analysis is a method of grouping customers by a shared starting event, usually their first order, and tracking them over the same elapsed time so groups compare fairly. A spring buyer is measured against a summer buyer at the same point in their lifecycle, never against last week's buyer.

A retention curve is the line you get by plotting the share of a cohort that orders in each period after their first purchase. Every curve falls. What matters is where it bends and where it settles.

Repeat purchase rate is the share of a cohort that has placed at least one more order by a fixed checkpoint after their first purchase. It should drive flow priorities, because each drop maps to a lifecycle moment you can change with email and SMS.

Most cohort content is written for subscription software, where a customer keeps paying until they cancel. In DTC ecommerce, every repeat order is a fresh decision, which is why the timing of the drop matters so much.

If you haven't pinned down where your second order is actually breaking down, start there, and anchor both reads in the metrics that actually matter for DTC retention.

Why does cutting cohorts by month hide the real story?

A monthly cohort blends customers who arrived for different reasons through different doors. Paid social impulse buyers, organic searchers, and deep-discount bargain hunters all land in one row, and their curves average into a smooth decline that describes none of them. You end up fixing an average customer who doesn't exist.

Month is the default cut in almost every tool, including Shopify's built-in cohort report, because it's easy to compute. It's also the least diagnostic cut you can make.

Say one month's cohort retains worse than the last. Was it the paid social campaign you scaled, the deeper welcome offer you tested, or a new hero product that disappoints on second use? The month view can't tell you, so teams guess, and the guess is usually "send more campaigns."

The blend can also hide problems entirely. A strong organic cohort can mask a collapsing paid social one, so a monthly line looks stable while one channel quietly leaks customers you paid to acquire.

Keep the monthly view for seasonality and trend. Just stop treating it as the diagnosis.

How should you cut cohorts instead of by month?

Cut every cohort three ways: by acquisition channel, by first product purchased, and by the offer used on the first order. Each cut isolates a different cause of churn. Channel tells you who you attracted, first product tells you what they experienced, and offer tells you why they bought in the first place.

Cut 1: Acquisition channel

Acquisition channel is the source that drove a customer's first order: organic search, direct, paid social, paid search, affiliate, influencer, referral, or owned email and SMS. According to Blossom's benchmark data, organic, direct, and referral customers typically carry the highest relative lifetime value, while paid social and affiliate buyers trend lower because they skew impulse- or deal-driven.

A paid social curve that falls faster isn't automatically a flow failure. It can mean the ad promised something the first box didn't deliver, which is a conversation with your paid team, not just a copy rewrite.

Cut 2: First product purchased

The entry SKU shapes the entire first experience. Some products are gateways to a second order; others are one-and-done. Cut this way, customer lifetime value (CLV) stops being one number and becomes a property of the product a customer started with.

A first product whose cohort bleeds points to a usage or expectation problem. A first product whose buyers reliably buy a specific second item hands you a cross-sell path to build into post-purchase.

Cut 3: First-order offer

First-order offer is the incentive attached to a customer's first purchase: usually the welcome discount from a popup, but also free shipping, a gift with purchase, or a sale price. In our experience with DTC clients, deep-discount cohorts retain worse than full-price ones, and according to Blossom's benchmark data, spin-to-win subscribers engage 20–40% less than standard popup subscribers.

If you aren't separating these, full-price buyers are propping up your blended curve, and you're overestimating how well discount cohorts stick. That's the case for converting subscribers without deep discounts.

Then read the cuts against each other:

• One cut diverges: the cause lives in that dimension, so fix it there.
• Every cut declines together: the problem is shared infrastructure, usually the post-purchase experience or fulfillment.
• Two cuts compound: a paid social, deep-discount row is often your worst. Fix the offer before rewriting the flow.

Keep each cell big enough to trust: according to Blossom's benchmark data, groups under about 500 customers produce metrics too noisy to act on, so merge small channels or widen the date window.

How do you read the shape of a retention curve?

Read the shape before the level. Where the curve bends tells you more than how high it sits. A sharp early cliff, a steady bleed, a plateau that suddenly breaks, and a curve that flattens onto a stable floor each come from different customer behavior and point to different fixes.

Plot each cut as the share of the cohort ordering in each period after their first purchase. The four shapes below are illustrative patterns, not one brand's data, but almost every real curve resembles one of them or a blend.

The cliff drop

The curve falls hard right after the first order, before customers would naturally need to reorder, and barely recovers. The first experience gave them no reason to return: the product didn't land, expectations were off, or nobody taught them how to use it. Discount and paid social cohorts show this most often.

The slow bleed

No dramatic break, just steady losses every period. Customers were satisfied but never built a habit, and the brand drifted out of mind. It's common when one-time buyers get the same campaigns as people who never purchased.

The plateau-then-drop

The curve holds through the first reorder or two, then falls at a specific point. Something expired: the product ran out with no reminder, a subscription incentive ended, or the winback flow fired after customers had already left.

The healthy asymptote

The curve drops early, as every brand loses its one-and-done buyers, then flattens onto a stable floor. That floor is your loyal core. The job here isn't repair; it's raising the floor and reaching it faster.

Reading these on live data is harder than it looks. Curves are noisy, cuts overlap, and seasonality fakes shapes: a Black Friday cohort full of sale buyers can look like a cliff drop even when your flows are fine.

Not sure which flow is causing your retention curve to drop? Get a free lifecycle audit and we'll show you exactly where the cohort is leaking.

Which lifecycle flow is responsible for each drop-off?

Match the drop-off's timing to the lifecycle stage that owns it. Early cliffs point to welcome and post-purchase, bleeds point to campaign relevance and cross-sell, plateau breaks point to replenishment and winback timing, and a flattening floor points to loyalty and VIP work. Timing, not size, picks the flow.

Post-purchase flow is the automated sequence after an order, built to reduce buyer's remorse, teach correct use, and open a path to the second purchase. Winback flow is the automated sequence that targets customers who have passed their normal repurchase window without ordering again.

Cliff drop: the curve-to-flow map

• Drop-off timing: Before the first natural repurchase window.
• Most likely flow: Post-purchase, plus the welcome flow that set expectations.
• Check first: Does post-purchase email one reassure, or does it pitch? Is there usage guidance before the product arrives?
• First fix to test: Confidence before monetization, then a cross-sell. For discount-heavy cohorts, rework the welcome sequence around turning new subscribers into faster first-time buyers with less offer dependency.

Slow bleed: the curve-to-flow map

• Drop-off timing: Gradual, every period after the first order.
• Most likely flow: Post-purchase cross-sell and campaign segmentation.
• Check first: Do one-time buyers get a separate campaign track from prospects?
• First fix to test: A product-specific cross-sell built from your first-product cut.

Plateau-then-drop: the curve-to-flow map

• Drop-off timing: Right after customers should have reordered.
• Most likely flow: Replenishment, then winback.
• Check first: Does the reminder land before or after the product runs out?
• First fix to test: Move reminders ahead of run-out; according to Blossom's benchmark data, a 30-day supply should trigger its first replenishment email around day 21–24. Then time fixing a lapsing cohort with a winback flow off your measured repurchase interval, not a guess.

Healthy asymptote: the curve-to-flow map

• Drop-off timing: Early loss, then flat.
• Most likely flow: Loyalty, VIP, and referral.
• Check first: Which channel, product, and offer make up the floor?
• First fix to test: Reward the floor with access rather than discounts by connecting your rewards program to email and SMS, and feed the floor's profile back to acquisition.

One sequencing rule: if a cohort shows both a cliff and a late drop, fix the cliff first. Customers who never reach the plateau can't be saved by winback.

What is a good retention curve for a DTC brand?

A good curve flattens onto a floor that meets your vertical's repeat-rate norm and reaches it fast. Benchmark the second order first, since that is where most cohorts leak, then compare your floor against brands that sell similar products at similar purchase frequencies, not against the ecommerce average.

First-to-second purchase rate: 20–30% is good, 30–40% is great, and 40%+ is elite, according to Blossom's benchmark data. Healthy time to second purchase runs 30–60 days according to Blossom's benchmark data, and under 21 days marks the elite tier.

Further out, according to Blossom's benchmark data, 90-day retention of 15–25% is good, 25–35% is great, and 35%+ is elite. Annual repeat rates also swing by vertical, according to Blossom's benchmark data:

• Health and supplements: 35–50%
• Beauty and skincare: 30–45%
• Fashion and apparel: 20–30%
• Food and beverage: 40–55%
• Home and lifestyle: 15–25%
• Pet: 40–55%

The most useful benchmark is internal. Your best-performing cut, say organic full-price buyers of your hero product, shows what your program does when the inputs are right. The gap between that row and your worst row is the size of the prize.

How do you build a cohort analysis in Klaviyo and Shopify without a data team?

You can build a working three-way cohort analysis with Shopify's order export, Klaviyo segments, and a spreadsheet. Tag each customer's first order with its channel, product, and offer, group customers by those tags, and count who orders again at fixed checkpoints. It takes an afternoon, not a data warehouse.

You don't need a BI tool to get a first read. Follow these steps in order:

1. Export your order history. Pull at least 12 months of orders using Shopify's order export, including customer email, order date, line items, discount codes, and landing site or referrer fields.
2. Isolate each customer's first order. Sort by customer and date, then flag the earliest order per customer. Every later order becomes a repeat event measured against that start date.
3. Label the three cuts. Add columns for channel (from UTM source or referrer), first product (the hero SKU on the first order), and offer. Group discount codes into tiers such as full price, light discount, deep discount, and free shipping so you aren't comparing hundreds of individual codes.
4. Set fixed checkpoints. For each customer, mark whether they placed a second order by day 30, 60, 90, and 180. Only count a checkpoint once the customer has been around long enough to reach it.
5. Pivot by cut. Build one pivot table per cut, with cut values as rows, checkpoints as columns, and the share of customers who reordered as values. Merge or drop any row below about 500 customers, the reliability threshold in Blossom's benchmark data.
6. Plot and classify the shape. Chart each row as a line and match it to the cliff, bleed, plateau-then-drop, or asymptote pattern before you look at absolute levels.
7. Mirror the cuts in Klaviyo. Sync first-order channel, product, and offer as profile properties, then build Klaviyo segments on them so each flow fix can target the exact cohort that's leaking.
8. Rerun on a schedule. Refresh the sheet monthly and review it properly each quarter, checking whether the fix you shipped actually changed the shape of the affected row.

Once the sheet exists, the hard part is discipline: change one flow at a time, note the date you shipped it, and read only the cohorts that entered after that date. Otherwise you'll credit a seasonal swing to a subject line.

Frequently asked questions

The short version: refresh cohort data monthly, review it quarterly, and judge each cohort only after its first natural repurchase window. Use cohorts to diagnose and RFM to target, widen cohorts if volume is thin, and keep subscription renewals on their own curve. The answers below expand on each point.

How often should you rerun a cohort analysis?

Refresh the data monthly and do a deeper review quarterly. Cohorts need time to mature, so a 90-day read on a cohort that started six weeks ago tells you nothing yet. Judge each fix only on cohorts that have had time to reach the checkpoint you care about.

How long should you wait before judging a new cohort?

Wait until the cohort has passed its first natural repurchase window. Judging earlier makes every cohort look like a cliff drop, because customers simply haven't needed to reorder yet. Always compare cohorts at the same elapsed time since first purchase.

What's the difference between cohort analysis and RFM segmentation?

Cohort analysis diagnoses where retention breaks; RFM targets who to message. Cohorts follow groups over time to show when customers stop reordering. RFM scores each customer's current recency, frequency, and monetary value. Use cohorts to decide which flow to fix and RFM to decide who receives it.

Can smaller brands use cohort analysis with limited order volume?

Yes, with adjustments. Widen cohorts from months to quarters, merge minor channels into an "other" bucket, and start with just one or two cuts instead of all three. The goal is cells large enough to show a real shape rather than random noise.

Should subscription orders be included in a repeat purchase curve?

Separate them. Auto-renewing subscription orders inflate repeat rates without reflecting a fresh buying decision, so a blended curve makes retention look healthier than it is. Treat subscribers as their own cohort, tracking skips and cancellations instead, and read one-time buyers on their own curve, where flows and offers actually change behavior.

A cohort chart only earns its place when it ends in a decision. Cut it three ways, read the shape, fix the flow that owns the drop, and measure whether the curve bends.

Want a second set of eyes on your curves? Book a free lifecycle audit and we'll map every drop-off in your cohorts to the flow that owns it.

Get retention teardowns like this in your inbox. Subscribe to our newsletter for practical lifecycle playbooks every month.