When repeat rate falls while acquisition looks healthy, the cause is almost always upstream. Either your blended repeat rate is being diluted by new customer volume and nothing is actually wrong, or the cohorts you bought 60 to 90 days ago are worse than the ones before them. Most brands go straight to the email program, which is almost never where it lives.
I’ve reviewed the numbers behind brands from $1M to $60M+, and this is the most misdiagnosed pattern in ecommerce optimization I run into. The founder sees the number slide, retention gets asked why, and three weeks later two extra campaigns a week are going to a list that was never the issue. The lag is what makes it hard: a customer on a 30-day supply doesn’t reorder until day 35 at the earliest, so a March acquisition decision doesn’t reach your retention reporting until late May.
Your blended repeat rate falls when you grow, even when every cohort is healthy
Most brands compute repeat rate as a trailing window: of everyone who ordered in the last 90 days, what share ordered more than once. That number moves when acquisition volume moves, because customers acquired last week have had no chance to reorder and sit in the denominator anyway.
Say every cohort behaves identically: 10% reorder by day 30, 18% by day 60, 25% by day 90. In a flat quarter at 1,000 new customers a month, that’s 250 repeaters from the oldest cohort, 180 from the middle, 100 from the newest. Blended, 530 of 3,000, or 17.7%.
Now you scale acquisition:
- January, 1,000 customers, 90 days elapsed: 250 repeaters
- February, 2,000 customers, 60 days elapsed: 360 repeaters
- March, 3,000 customers, 30 days elapsed: 300 repeaters
- Blended: 910 of 6,000, or 15.2%
Your blended repeat rate dropped 2.5 points and cohort behavior didn’t change by a hundredth of a point. The faster you grow, the worse the number looks. Something like a third of the panic cases I get asked about resolve right here.
The cohort view that replaces the blended number
Rebuild it by first-order month and measure at a fixed horizon.
Repeat purchase rate: Customers with 2 or more orders ÷ total customers in the window
Cohort repeat rate at day N: Customers in a first-order cohort with a second order within N days of their first ÷ total customers in that cohort
Time between orders: Median days from first order to second order, calculated only on customers who placed a second order
Cohort by the month of a customer’s first order, never the month of the order itself. Hold the horizon at day 60 and day 90, and drop any cohort that hasn’t matured past it.
Day 60 is what I watch weekly, day 90 is what I report. Time between orders is the leading indicator underneath both: when the median stretches from 41 days to 52, day-60 repeat falls next month. This sits inside the ecommerce metrics that predict scale.
The upstream-first diagnostic sequence
Run these gates in order. Each is cheaper to check than the next, and fixing a downstream symptom while the upstream cause keeps running wastes a quarter.
Gate 1: Is it mix, or is it real?
Compare cohort repeat rate at day 90 across the last six mature cohorts. If they all sit inside a two-point band, the decline is a mix artifact and you’re done. If one broke from the pattern, note the month. Whatever you changed 30 days before it is your suspect list.
Gate 2: Is it a cohort-quality problem?
This is where the answer lives most of the time. Split the broken cohort by first-touch channel, first-order discount rate, and entry offer, then compare each slice against the same slice in a healthy cohort. Look for a mix shift toward broad prospecting, a discount-led offer that bought price shoppers, a promotion that pulled in buyers with no product intent, or a creative angle attracting someone your product doesn’t retain. A 40% off sitewide push beats every CPA target you have and hands you a cohort that reorders at half your rate.
Gate 3: Is it a product-experience problem?
If cohort composition matches your healthy cohorts, look at what the customer received. Formulation changes, a supply substitution nobody flagged as customer-facing, a stockout on the hero SKU during the reorder window, slower shipping, a fulfillment partner change. Pull median order-to-delivery days and out-of-stock days for the reorder SKU by month, next to the cohort curve.
Gate 4: Is it a lifecycle problem?
Now you can look at email and SMS, and you’re looking for breakage before strategy. Flow entry counts by day, inbox placement, whether cadence matches the reorder window, whether a platform migration silently killed a trigger. A replenishment flow that stopped firing on the fifteenth leaves a visible signature.
Gate 5: Is it a pricing or cadence problem?
Price increases, pack size changes, and subscription default changes at checkout all move repeat rate with a delay. Going from 30 servings to 45 pushes the reorder window out two weeks, so day-60 repeat falls even though annual customer value went up. Adjust the horizon for that one rather than treating it as a problem.
| Cause | Signal that confirms it | Lag | Fix |
|---|---|---|---|
| Channel mix shift to prospecting | Repeat by first-touch channel diverges 5+ points | 60-90 days | Cap spend in the weakest-repeating channel |
| Discount-led offer or promotion | First-order discount rate up 10+ points | 60-90 days | Shift the offer to bundle value or subscription |
| New angle, different buyer | Product mix and AOV shift inside the cohort | 45-90 days | Judge angles on day-60 repeat, not day-1 CPA |
| Formulation or supply substitution | Tickets on taste, texture, smell; rating dip | 30-60 days | Tell customers before it ships |
| Stockout on the reorder SKU | Out-of-stock days overlap the reorder window | 30-45 days | Back-in-stock flow, hero SKU safety stock |
| Shipping time or 3PL change | Median order-to-delivery up 2+ days | 30-60 days | Service-level fix, then delay notices |
| Deliverability drop | Flow revenue per recipient falls, sends flat | 15-45 days | Prune, warm back, split promo and flow sending |
| Trigger broken by a migration | Flow entries per day drop toward zero | Found late | Weekly flow-entry audit with an alert |
| Price, pack size, subscription default | Time between orders stretches, repeat holds | 30-60 days | Adjust the horizon before calling it a decline |
The deliverability check almost nobody runs
Deliverability hides, because everything on the surface still looks normal. Sends go out and campaign reports populate. Open rate has been unreliable since Apple started prefetching, so the metric that would have caught it early is the one you stopped trusting.
A big promotional send to a poorly pruned list drags domain reputation down, inbox placement slides from the low nineties into the sixties, and your flows keep firing into spam folders. If flows and campaigns carry 40% of your revenue and half of that lands where the customer never sees it, you just lost a fifth of your retention revenue without a single dashboard changing color.
The check takes twenty minutes. Pull revenue per recipient for core flows by week over the last twelve weeks, holding send volume constant. Flow audiences are stable, which makes flow performance the cleanest deliverability signal you have. A 30% drop across every flow in the same week is a delivery problem. Then check complaint rate, bounce rate, and whether a big campaign preceded it by seven to ten days. Where flows fit in a full program is covered in customer loyalty campaigns for supplement brands.
What healthy looks like, and what counts as noise
Ranges I use for supplement and wellness DTC, measured by cohort:
Day 60 repeat rate: 12-20% is normal. Above 22% is a strong product-market fit signal.
Day 90 repeat rate: 20-30% is workable. Above 35% means you can outspend most competitors. Under 15% is a product or expectation problem no budget fixes.
Time between first and second order: 35 to 55 days on a 30-day supply. Past 65 days, your replenishment timing is wrong.
Movement of 1 to 2 points cohort over cohort is noise, and I ignore single-cohort moves entirely. Two consecutive cohorts down 3 or more points is signal, and you should be at Gate 2 that week. Subscription retention runs on its own curve, covered in subscription retention benchmarks for supplements.
What to do in the first two weeks
- Days 1-2. Rebuild the cohort table by first-order month, day 60 and day 90, six mature cohorts. If they’re inside a two-point band, stop here.
- Days 3-4. Run the deliverability check and audit flow entry counts. It either clears email or convicts it.
- Days 5-7. Split the broken cohort by channel, discount rate, and entry offer against a healthy one. Most answers surface here.
- Days 8-10. Pull operations for that window: out-of-stock days, delivery time, supplier or 3PL changes, ticket themes.
- Days 11-14. Change one upstream thing and write down the measurement date, since an offer or channel fix won’t report back for 60 days.
Don’t add campaign volume during this window. More email to a list that isn’t the problem contaminates the only clean signal you have. To recover a cohort that already lapsed, run the win-back sequence for lapsed subscribers.
When a falling repeat rate is the right answer
Sometimes the number should go down and you should let it.
Open a new top-of-funnel channel and blended cohort quality dilutes by design. The question is whether the incremental volume is accretive on contribution margin at a 90-day horizon. A cohort repeating at 18% that costs 40% less to acquire can be a better business than one repeating at 26%.
Same with a trial offer or a sample-size SKU. You bought a wider top of funnel at a lower first-order value and a lower expected repeat rate, and if you priced it right, that’s the plan working. Judge it on payback period and 90-day contribution margin per customer. Broad prospecting after retargeting-heavy spend behaves the same way, since retargeting cohorts always look better on repeat.
The test is whether you predicted it. A decline that surprised you is a diagnosis you owe yourself.
Common questions
What is a good repeat purchase rate? For supplement and wellness DTC, measured by cohort: 12-20% at day 60 and 20-30% at day 90. Above 35% at day 90 is a strong business. Under 15% points to a product or expectation problem no budget fixes.
Why did my repeat rate drop after a good month? A good acquisition month adds customers to the denominator who haven’t had time to reorder, so a trailing-window calculation falls even when every cohort is healthy. Rebuild by first-order cohort at a fixed horizon first.
How do I run a cohort analysis without a data team? Export orders with customer ID and order date. Add a column for each customer’s first order date, one for days since it, and a flag for whether a second order landed within 60 and 90 days. Pivot with first-order month as rows.
How long before an acquisition change shows up in retention? 60 to 90 days on a 30-day supply, closer to 120 days on a 60-day supply. By the time the number moves, the decision behind it has been running for two months, which is why the fix belongs upstream.
Is my email program the problem? Usually not, though it can amplify a problem that started upstream. Check flow entry counts per day and revenue per recipient on core flows over twelve weeks. If entries are steady and revenue per recipient is flat, email is doing its job.
If your repeat rate has been sliding for two months and nobody can tell you which cohort broke, that’s a measurement problem before it’s a retention problem. Bring twelve months of orders with customer IDs and spend by channel, and we can find the cohort and the decision behind it in one session. Here’s how engagements work if you’d rather read first.