A blended repeat rate is the most flattering number in a D2C dashboard, and it is flattering for a structural reason: it mixes customers who have had three years to reorder with customers who have had three weeks. A D2C cohort retention analysis removes that distortion by grouping customers according to when they first bought and then tracking each group separately over time. For brands selling replenishable products (coffee, supplements, pet food, skincare, razors, cleaning goods), the cohort table is the only report that shows whether new customers are actually behaving like the old ones, and whether the reorder curve that justified last quarter’s acquisition budget still holds.
In short
- Blended repeat rate averages old and new customers together, so a brand can look healthier every month while each new cohort quietly retains worse than the one before it.
- A cohort table needs only three columns of raw order data (customer ID, order date, order value) and can be built in a spreadsheet before any analytics tool is involved.
- For replenishable products the key shape is the reorder curve: the share of a cohort that places a second order by day 30, 60, 90 and 180, read against the product’s natural consumption window.
- Segmenting cohorts by acquisition channel and first product usually explains more variance in retention than any change in email or loyalty tactics.
- The report earns its place when it changes three decisions: how much to pay for a new customer, which first products to push, and how much of next quarter’s revenue is already owned by existing cohorts.
Why does blended repeat rate mislead growing brands?
Blended repeat rate is calculated as the share of customers (or orders) in a period that came from someone who had bought before. The problem is the denominator: it includes every customer the brand has ever acquired, regardless of how long they have had to reorder. A brand that acquired 5,000 customers in 2024 and 20,000 in 2026 is mostly measuring 2024’s survivors against 2026’s newcomers, and the two groups have nothing in common except the product.
Growth makes the distortion worse rather than better. When acquisition accelerates, the customer base skews toward people who have not yet had time to place a second order, which drags blended repeat rate down even if every cohort is retaining perfectly. When acquisition slows, the opposite happens: the base ages, the survivors dominate, and repeat rate rises while the business is actually shrinking. Founders working through the numbers in the honest version of scaling D2C from one million to ten million tend to hit this exact trap at the point where paid acquisition doubles.
There is a second, quieter failure. Blended repeat rate cannot show timing. A replenishable product has a natural consumption cycle: a 30-day supplement, a 12-week coffee subscription, a 6-month skincare regimen. A customer who reorders at day 35 and a customer who reorders at day 190 both count as “repeat” in a blended metric, yet they represent completely different economics, inventory plans and marketing calendars.
The broader picture of how funding, founders and exits interact with retention numbers is covered in the retail business landscape guide; this article stays inside the cohort table itself.
What a cohort actually is
A cohort is a group of customers who share a starting point, almost always the month (or week) of their first order. Every cohort is then observed at fixed intervals after that starting point: month 1, month 2, month 3, and so on. Because each cohort is aligned to its own start date, a 2024 cohort at month 6 is directly comparable with a 2026 cohort at month 6. That alignment is the entire value of the method, and it is why analysts across finance, medicine and product management use the same structure (Wikipedia’s overview of cohort analysis traces the shared origins).
The three questions a cohort table answers that repeat rate cannot
- Are customers acquired this quarter retaining as well as customers acquired a year ago, measured at the same age?
- When does a typical customer place their second, third and fourth order, and is that window moving?
- How much revenue will existing customers generate next quarter if the brand acquires nobody new?
How do you build a first cohort table from raw order data?
A working cohort table needs three fields per order: a stable customer identifier, an order date, and an order value net of refunds and discounts. Shopify, WooCommerce, BigCommerce and every order management system export these. Everything else (product, channel, discount code) is useful later for segmentation but is not required for the first pass. The first table should be built in a spreadsheet, not a BI tool, because the founder needs to see the mechanics before trusting a dashboard that hides them.
Step by step
- Export every order since launch with customer ID, order date and net value. Exclude cancelled and fully refunded orders; keep partial refunds at their net value.
- Find each customer’s first order date and assign the customer to a cohort labeled by that month (for example, 2026-03).
- Compute the age of every order as the number of months between the order date and the customer’s first order date. A second order 45 days later is month 1; a fourth order 200 days later is month 6.
- Count, per cohort and per age month, how many distinct customers placed at least one order. That count divided by the cohort’s starting size is the retention percentage.
- Repeat the same grid for revenue, summing net order value per cohort per age month, and divide by the cohort’s starting size to get revenue per original customer.
The result is a triangle, not a rectangle. The oldest cohorts have data at month 12 or beyond; the newest cohort has data only at month 0. Cells to the right of each cohort’s current age are blank, and they should stay blank. Filling them with projections inside the same table is the most common way to lose the distinction between what happened and what was forecast.
Choosing the interval: weekly, monthly or quarterly
Monthly cohorts are the default for most replenishable brands because the reorder cycle is measured in weeks and the sample per cohort stays large enough to be meaningful. Weekly cohorts suit brands running frequent promotions, where a single discount event can define a cohort’s quality. Quarterly cohorts are only useful for very low-volume brands where a monthly cohort would contain fewer than 100 customers, at which point retention percentages become noisy enough to mislead.
| Cohort (first order month) | Customers | Month 1 | Month 2 | Month 3 | Month 6 |
|---|---|---|---|---|---|
| 2026-01 | 1,200 | 34% | 41% | 46% | 52% |
| 2026-02 | 1,450 | 33% | 40% | 45% | 50% |
| 2026-03 | 2,900 | 26% | 32% | 37% | (not yet) |
| 2026-04 | 3,100 | 24% | 30% | (not yet) | (not yet) |
| 2026-05 | 2,700 | 27% | (not yet) | (not yet) | (not yet) |
The figures above are an illustrative table for a 30-day consumable, not a benchmark. Each cell shows the share of the cohort that had placed at least one additional order by that age. Read it and the story is immediate: the January and February cohorts are on the same curve, March doubled in size and dropped roughly eight points at every age, and April continued the slide. A blended repeat rate over the same period would have risen, because the large early-2026 cohorts were entering their reorder window at the same time.
How do you read the reorder curve for a replenishable product?
The reorder curve is a single cohort’s retention row plotted against age. For a replenishable product it has a characteristic shape: a flat stretch during the consumption period, a steep rise as the first pack runs out, then a slower climb as late reorderers trickle in. The three things to read from it are the height of the first step, the timing of that step relative to the product’s consumption window, and the slope after the step.
Cumulative versus active retention
Two definitions of “retained” are in circulation and they produce very different curves. Cumulative retention (sometimes called “ever repeated”) counts a customer as retained at month N if they have placed any order since their first, so the curve can only rise. Active retention counts a customer as retained at month N only if they ordered in that specific month, so the curve typically falls and then stabilizes. Replenishable brands should track both: cumulative retention shows how many customers ever came back; active retention shows how many are still consuming at the expected cadence.
The mistake to avoid is comparing a cumulative curve from one report with an active curve from another. A brand that reports “60% retention at month 6” using cumulative math and “22% retention at month 6” using active math is describing the same cohort, and both numbers are correct. The unit economics section of the D2C unit economics every founder should be able to defend depends on the active definition, because lifetime value is a sum of what customers actually spend in each period.
What the first step tells you
The height of the first reorder step (the share of a cohort that reorders within one consumption cycle) is the single most predictive number in the table. Across replenishable categories, a cohort’s eventual month-12 cumulative retention is usually a fairly stable multiple of its first-cycle reorder rate, which means a brand can grade a cohort’s quality within 45 to 60 days of acquiring it rather than waiting a year. Once a brand has six or more mature cohorts, it can calculate its own multiple and use it to forecast new cohorts early.
Product-specific curve shapes
Coffee and pet food produce tight, early steps because consumption is daily and the pack size is predictable. Supplements produce a softer step because adherence varies, and a meaningful share of customers stop taking the product before the pack is empty. Skincare produces the latest and widest step, since a single unit can last two to four months and the customer may not notice depletion until well after. Knowing the expected shape for the category prevents a brand from panicking at a month-1 retention figure that is normal for its product.
What is an expected reorder window and what does a late cohort mean?
An expected reorder window is the range of days after first purchase during which a satisfied customer should logically run out and reorder. It is set by the product, not by the marketing team: a 60-capsule supplement at two per day has a 30-day window with a tolerance of perhaps 10 days either side. Defining the window explicitly turns retention from a vague percentage into a testable expectation. Customers who reorder inside the window are behaving as designed; customers who reorder late are signaling reduced usage; customers who never reorder are churned.
Deriving the window from the data rather than the label
The label says 30 days; the data usually says something else. Plotting the distribution of days between first and second order for mature cohorts reveals the actual median reorder gap, and it is common to find that the real median sits 20% to 40% beyond the label because customers skip doses, share packs or stockpile from a launch promotion. The data-derived window is the one to use in the cohort table, because it reflects how customers actually consume the product.
Three readings of a late cohort
- The cohort reorders late but eventually catches up. This usually means the first order was larger than normal (a bundle or a promotional multipack) and the cohort simply had more product to get through. Check average first-order units before diagnosing a retention problem.
- The cohort reorders late and never catches up. This is a quality signal: the cohort was acquired with weaker intent, often through a heavy discount or an audience that does not use the product daily. Trace it back to channel and creative.
- The cohort reorders early. Rare but real, and it can be a warning: customers ordering before they should run out are often taking advantage of a second-order discount and will show up as churned at month 3 once the discount is gone.
Brands that added a subscription option often see the window compress artificially, because the subscription bills on schedule regardless of consumption. That is not retention in the behavioral sense, and the cohort analysis for subscribers has to be run separately, with churn defined as cancellation rather than non-reorder. The category-level picture of where subscriptions actually work is laid out in the analysis of which D2C subscription categories hold up.
How should you segment cohorts by acquisition channel and first product?
A single cohort table for the whole brand hides the fact that customers acquired through different channels and different entry products behave like different businesses. The first segmentation to run is by acquisition channel (paid social, paid search, organic, affiliate, marketplace, referral), because that is where the budget decision sits. The second is by first product, because the entry SKU shapes the reorder window and the probability of a second order more than any downstream tactic.
Channel cohorts
The recurring pattern across replenishable brands is that paid social cohorts show the steepest early step and the fastest decay, organic and referral cohorts show a lower first step but a longer tail, and marketplace cohorts (Amazon, Walmart) are often untrackable at the customer level and have to be excluded rather than estimated. Once channel cohorts are separated, the brand can compute a channel-specific customer lifetime value and compare it directly with the channel’s acquisition cost, which is the calculation every board actually wants to see.
Attribution matters here, and it is imperfect. A customer who clicked a paid ad after seeing a creator video will be logged as paid social. The analysis is still valuable as long as the attribution method is consistent across cohorts, because the comparison between cohorts is what drives decisions, not the absolute label. Changing the attribution model mid-year breaks cohort comparability and should be flagged in the report when it happens.
First-product cohorts
For a brand with a hero product and a long tail, first-product cohorts almost always show that the hero produces the best retention and the discounted starter kit produces the worst. The surprising finding tends to be in the middle: a secondary product that quietly retains as well as the hero, or a bundle that retains well but at a much lower revenue per customer because it front-loads three months of supply. These findings feed directly into which products get the acquisition budget and which get retired from paid campaigns.
| Segment dimension | Best used for | Typical finding in replenishables | Decision it changes |
|---|---|---|---|
| Acquisition channel | Setting CAC ceilings per channel | Paid social retains worse than organic at month 6 | Budget allocation and bid caps |
| First product | Choosing entry SKUs for campaigns | Hero SKU outperforms starter kits and bundles | Which products get paid traffic |
| Discount at first order | Testing promotion depth | Cohorts acquired above 25% off decay fastest | Maximum first-order discount |
| Geography or market | International expansion tracking | New markets retain lower until local fulfillment exists | Whether to fund local stock |
| Subscription vs one-time | Separating billed from behavioral retention | Subscriber “retention” masks passive churn | Cancellation flow and dunning |
Geography as a segment
Brands expanding beyond their home market should treat each country as its own cohort set from day one, because shipping times, duties and local competition change the reorder window and the second-order probability. A 30-day consumable delivered in 10 days behaves like a 20-day consumable, and customers who face customs charges at the door churn at a rate that has nothing to do with the product. The margin implications are worked through in the guide to scaling D2C internationally without losing your margin, and cohort tables are the instrument that shows whether a new market is retaining well enough to justify local inventory.
How do you forecast revenue from existing cohorts?
The most underused output of a cohort table is the revenue that existing customers will generate with zero new acquisition. It is calculated by taking each live cohort, reading its current age, and applying the revenue-per-customer curve observed in mature cohorts for the ages still ahead. Summed across cohorts, that gives a floor for next quarter’s revenue that is already “owned,” and the gap between that floor and the revenue target is the amount acquisition actually has to deliver.
The mechanics
- Build the revenue-per-original-customer grid (net revenue per cohort per age month divided by cohort size).
- For mature cohorts (those that have reached the forecast horizon), average the revenue per customer at each age to form a reference curve. Weight recent cohorts more heavily if the curve is drifting.
- For each live cohort, multiply its original size by the reference curve values at the ages that fall within the forecast period.
- Sum across cohorts. This is the owned revenue for the period.
- Subtract from the revenue plan. The remainder is the new-customer revenue required, which, divided by expected first-order value and first-period repeat, gives the number of new customers acquisition has to produce.
Worked through for the illustrative table above, a brand with 11,000 customers acquired in the first five months of 2026 and a reference curve worth roughly $18 per original customer across months 3 through 5 would carry about $200,000 of owned revenue into the following quarter. That number is not a forecast in the speculative sense; it is what those customers will spend if they keep behaving the way earlier cohorts did, which makes it the most defensible line in a board deck.
Where the forecast breaks
The reference curve assumes the newest cohorts will follow the mature ones. When the cohort table already shows a decline in early-age retention (as the March and April cohorts did in the example), the reference curve overstates owned revenue and the forecast should be built from the most recent comparable cohorts instead. Seasonality is the second break point: a cohort acquired in a gifting season often includes buyers who were never going to reorder, and its curve should be compared with the same season a year earlier rather than with the adjacent months.
Price changes are the third. A brand that raised prices 12% in June will see revenue per customer rise for every cohort at every subsequent age, which looks like improved retention on the revenue grid while the customer-count grid is unchanged or worse. Reporting both grids side by side is the safeguard. The mechanism connecting those two grids to gross margin and contribution is the same one the unit economics guide uses to build a defensible lifetime value.
What are the three decisions this report should change?
A cohort retention report that does not change a decision is a chart, not a report. Three decisions sit closest to the numbers, and a brand should be able to point to the cell in the table that drove each one.
1. How much to pay for a new customer, by channel
Channel-level cohort curves turn the customer acquisition cost debate from an argument about opinions into a calculation. If paid social cohorts generate $54 of contribution margin over 12 months and paid search cohorts generate $71, the brand has a factual basis for a different bid ceiling in each channel. The number to use is contribution margin, not revenue, and the horizon should match the payback period the business can afford, which for a bootstrapped brand is rarely longer than six months.
2. Which first product to put in front of new customers
First-product cohorts identify the entry SKU with the best combination of second-order probability and revenue per customer. Very often it is not the cheapest product and not the most popular one either; it is the product whose consumption cycle is most predictable. Once identified, that product should receive the acquisition creative, the landing page investment and the sampling budget, while low-retention entry products are moved out of paid campaigns even if they convert well on the first click.
3. How much of next quarter’s revenue is already owned
The owned-revenue calculation sets the acquisition target and, indirectly, the cash plan. A brand that knows 55% of next quarter’s revenue is already owned can afford to hold acquisition spend steady while it fixes a retention problem. A brand at 25% has no such buffer and needs acquisition to perform every week. Finance teams that have this number stop asking for a blended repeat rate, because owned revenue answers the question they were actually asking.
Loyalty and email programs are the levers most brands reach for first when the owned number is low; how those programs earn repeat sales, rather than simply reward them, is examined in the guide to designing a loyalty program that earns repeat sales.
Common mistakes that make the report unusable
- Mixing gross and net order value. Refunds on replenishables are small per order but concentrated in the worst cohorts, so gross figures flatter exactly the cohorts that need scrutiny.
- Counting subscription renewals as reorders. A renewal is a billing event. It should be tracked in a separate subscriber cohort with cancellation as the churn event.
- Merging customer records inconsistently. Guest checkouts and account checkouts by the same person split one customer into two, inflating cohort sizes and depressing retention. Deduplicate on email before building the table.
- Comparing cohorts at different ages. Month 3 for the new cohort versus month 9 for the old one tells you nothing. Compare like ages only.
- Projecting into the triangle’s blank cells. Keep observed data and forecasts in separate tables so nobody presents a projection as a result.
- Updating the table quarterly. For a 30-day consumable, a quarterly refresh means the brand learns about a bad cohort three cycles after it could have acted. Monthly is the minimum; weekly for high-volume brands.
FAQ on cohort retention reporting
What is a D2C cohort retention analysis?
A D2C cohort retention analysis groups customers by the month of their first purchase and tracks what share of each group places further orders at fixed ages afterward (month 1, month 2, month 6, and so on). Because every cohort is aligned to its own start date, the method compares customers of the same age rather than mixing new and old buyers together. For replenishable products it shows the reorder curve, the expected reorder window, and how much revenue existing customers will produce without any new acquisition.
How is cohort retention different from repeat purchase rate?
Repeat purchase rate is a blended figure: the share of customers or orders in a period that came from returning buyers, regardless of how long those buyers have been customers. Cohort retention holds customer age constant, so a cohort acquired in January is only ever compared with another cohort at the same age. Repeat rate rises when acquisition slows and falls when it accelerates, which makes it a poor guide for a growing brand; cohort retention is unaffected by the mix of ages in the base.
What is a good month-1 retention rate for a replenishable product?
There is no universal benchmark, because month-1 retention depends on the consumption window. A 30-day consumable should show most of its first reorder step by month 1 or 2, while a three-month skincare product may show almost nothing until month 3. The useful comparison is internal: each new cohort against the brand’s own mature cohorts at the same age. A cohort that lands more than a few points below the brand’s established curve at the first reorder step is the one to investigate.
Should cohorts be built on customers or on orders?
Customers. An order-based cohort double counts heavy buyers and hides the fact that a shrinking number of people are producing the same order volume. Build the customer-count grid first (distinct customers active per cohort per age), then the revenue grid (net revenue per original customer). Reading both together separates a retention change from a price or basket-size change, which order-based tables cannot do.
How do subscriptions fit into a cohort retention table?
Subscribers should be a separate cohort set. A subscription renewal is a billing event that happens whether or not the customer is consuming the product, so it does not carry the same behavioral signal as a voluntary reorder. For subscriber cohorts, the churn event is cancellation (or a failed payment that is not recovered), and the useful curve is the share of each signup cohort still active at each age. Mixing subscribers with one-time buyers inflates early retention and hides passive churn.
How much order history is needed before the table is useful?
At least two full reorder cycles of history, so the oldest cohorts have reached an age where the first reorder step is complete. For a 30-day consumable that is roughly three to four months of data; for a quarterly product it is closer to nine months. Cohorts need a minimum size too: below about 100 customers, a single cohort’s retention percentages swing enough from random variation to mislead, and weekly cohorts should be rolled up into monthly ones.
Which tools can build a cohort table?
A spreadsheet is enough for the first table and is the recommended starting point, because the founder sees every step. Shopify’s built-in analytics, Google Analytics 4, and specialist retention platforms all produce cohort views, but their definitions of “retained” differ (cumulative versus active, orders versus customers, gross versus net), so figures from different tools are rarely comparable. Whatever tool is used, the definitions should be written into the report so a reader can reconcile them.
How do you forecast revenue from a cohort table?
Take the revenue-per-original-customer curve from mature cohorts as a reference, then apply it to each live cohort for the ages that fall inside the forecast period, multiplying by the cohort’s original size. Summing across cohorts gives the revenue existing customers are expected to produce with no new acquisition. The difference between that figure and the revenue plan is what new customers must deliver, which converts directly into an acquisition target for the quarter.
What is the biggest mistake in cohort reporting?
Comparing cohorts at different ages. A new cohort at month 2 will always look worse than an old cohort at month 9, and it says nothing about quality. The second most common mistake is failing to deduplicate customer records, which splits one buyer across guest and account checkouts and depresses every retention figure. Both errors are structural and will survive any amount of dashboard polish; they have to be fixed in the data preparation step.
What to read next
Cohort retention is the reporting layer; the decisions it feeds sit in the economics. The full method for turning retention curves into a defensible lifetime value and payback period is in the D2C unit economics guide, and the wider context of what investors and acquirers read into those curves is set out in the retail business landscape: funding, founders and exits.