Most operators report retention as one number. “We retain about 28 percent.” That figure is a blended average, and the average is exactly the thing that lies to you. Two brands can post the same headline rate while running completely different businesses underneath, one quietly heading to zero, one compounding. The difference is not in the number. It is in the shape of the curve.
A retention curve plots the share of a single acquisition cohort that is still active, measured at each period after their first purchase. You fix the cohort (say, everyone acquired in January), then track what fraction return in month one, month two, and so on. The standard literature on cohort analysis is clear that the shape, not any single point, is what tells you whether a product is sticky (Userpilot, Retention Curve guide, 2024; Churnkey, Customer Retention Curves, 2024).
There are three shapes worth knowing by name.
Three canonical shapes
Plot enough cohorts and they almost always fall into one of three patterns. Read left to right, and the question you are answering is simple: where does the curve go after the early drop?
Declining: no floor, no fit
The declining curve slopes toward zero and never settles. Every cohort eventually churns out entirely. This is the signature of a business with no product-market fit, where growth is purely a function of how fast you can buy new customers. It can look healthy on a revenue chart for a long time, because fresh acquisition papers over the leak. The cohort view is what exposes it.
Flattening: the healthy default
The flattening curve is the most common shape for a healthy direct-to-consumer or service business. There is a steep early drop, then the curve levels off and holds. That floor is the single most useful number in the chart: it is your loyal-customer rate, the share of any cohort that becomes durable. A flattening curve is one of the strongest signals a product can have, because it means a segment has formed a habit and keeps finding value. The higher and earlier the curve flattens, the healthier the cohort.
Smiling: the rare one
The smiling curve drops, stabilises, and then turns back up. Net revenue or active users in a cohort actually grow over time as lapsed buyers reactivate or existing accounts expand faster than others churn. This is the pattern behind the best collaboration and network-effect products, where a cohort becomes more valuable the longer it exists. It is rare, and when you see it you protect it.
The headline repeat rate tells you the height of the first step. The shape tells you whether the business compounds. Only one of those is worth managing.
Why the average misleads
Here is the trap in concrete terms. Across 156,110 direct-to-consumer customers, the measured repeat-purchase rate was 18.8 percent, meaning 81.2 percent bought exactly once and never came back (BS&Co, Repeat Purchase Rate Benchmarks, 2024). That is well below the 25 to 30 percent figure many operators quote as the DTC “average.” The gap is the lesson: blended averages mask that most stores live in the high teens to low twenties, and a flattering benchmark can hide a declining curve. The pattern holds for Australian retailers. Australia Post's eCommerce Report 2026 finds online purchase frequency rising, with nearly half of Millennials now buying online weekly or more, yet the same structural truth applies here: a single repeat rate averaged across a base hides which cohorts actually come back.
Category dispersion makes it worse. Repeat behaviour is structural, not just a marketing outcome. Replenishment categories retain far better than considered, infrequent ones.
Category repeat-rate ranges per Finsi.ai ecommerce retention benchmarks and Mageloyalty 2026. Curve tendency is interpretive.
The decay across a one-off purchase cohort is steep: with most categories settling into the high teens to low thirties on an annual repeat basis (Finsi.ai, repeat-purchase benchmarks), the floor sits far below where subscription cohorts hold. A subscription comparison sets the bar for the best case: across 1,200-plus subscription sites, average monthly churn was 3.27 percent, with about 4 percent considered good (Recurly Research, Churn Rate Benchmarks, 2023 study period). A 4 percent monthly churn implies roughly 61 percent of a cohort still active after twelve months, the kind of high, flat floor that one-off purchase brands rarely reach.
The math: fit the curve, do not eyeball it
Eyeballing the shape is fine for a first read. To compare cohorts and to project the floor, fit a curve. The standard practitioner choice is a power law, because retention decays fast then slows in a way that a straight exponential overstates.
This is textbook cohort-fitting, the same approach used in public retention calculators (miniwebtool, Cohort Retention Calculator). Once you have a and k, you read the steady-state floor by extrapolating where the curve levels, and you have a clean, comparable number per cohort rather than a noisy month-by-month series.
Worked example
Take a flattening cohort with retention of 42 percent at month 1, 28 percent at month 3, and 24 percent at month 6. Regress log R on log t. The two-point slope between month 1 and month 6 gives the decay exponent directly: k = −(log 0.24 − log 0.42) / (log 6 − log 1).
A k near 0.3 is a shallow decay: the cohort is settling toward a real floor around the high teens, not sliding to zero. Run the same fit on a declining cohort and k comes back far higher, with no stable floor to extrapolate. That single exponent, compared across acquisition months and channels, tells you which cohorts are worth acquiring more of.
Where the curve bends: the second order
The early steep section is not destiny. The inflection that decides whether a curve declines or flattens is the second purchase. From the 156K-customer dataset, 50.3 percent of repeat buyers made their second purchase within 30 days and 76.4 percent within 90 days (BS&Co, 2024). And momentum compounds: the foundational RJMetrics benchmark of 176 retailers and 18 million customers found that after a first order a customer has only a 32 percent chance of a second, but once they place that second order the chance of a third climbs to 54 percent (RJMetrics, Ecommerce Buyer Behavior, 2015). The conditional probability of the next order keeps rising at each step. Plan post-purchase flows on the median time-to-second-order, around 15 to 35 days, not the long-tail mean.
The economic stakes are well documented. Bain & Company found that increasing retention by 5 percent can raise profit by 25 to 95 percent, and that acquiring a new customer costs five to twenty-five times more than retaining one (Reichheld & Bain, Prescription for Cutting Costs). Lifting the flatten point of your curve, not just the height of step one, is what moves those numbers.
For Auto Comfort, Blufire worked the second-conversion inflection rather than raw lead volume: a 4.84:1 return and a $41.70 cost per enquiry, by lifting the share of enquiries that became jobs instead of buying more enquiries.
What this is in Profit Velocity terms
Blufire measures businesses on Profit Velocity, the rate at which marketing and sales effort converts into durable contribution margin. A flattening curve with a high floor is Profit Velocity made visible: the numerator, durable contribution margin, compounds and persists rather than resetting every month. Raising the flatten point lifts lifetime value and cuts churn at the same time, which is the cleanest lever there is. A declining curve is the opposite. It tells you that revenue depends on ever-rising acquisition spend, so the denominator keeps growing.
How to use it on Monday
- Stop reporting one repeat rate. Report the curve by acquisition cohort and read its shape first.
- Fit k per cohort and per channel. The decay exponent makes cohorts comparable and exposes channels that buy one-and-done buyers.
- Find your flatten point. The floor is your loyal-customer rate; it is the number to grow, not the month-1 step.
- Instrument the second order. Build post-purchase flows around the 15-to-35-day median window, because that is where the curve bends.
A retention curve is the cheapest diagnostic you own. It needs no new data, just your existing orders grouped by cohort. Read the shape, fit the decay, and you will know in twelve months of history whether you are building something that compounds or something you have to keep refilling.
Primary sources
- BS&Co, “Repeat Purchase Rate Benchmarks: 18.8% Across 156K Customers,” 2024. bsandco.us
- Recurly Research, “Customer Churn Benchmarks,” 1,200+ subscription sites, 2023 study period. recurly.com
- Reichheld & Sasser / Bain & Company, “Prescription for Cutting Costs” (5% retention to 25-95% profit; 5-25x acquisition cost). bain.com
- Userpilot, “Retention Curve: How to Measure and Improve It,” 2024. userpilot.com
- Churnkey, “Customer Retention Curve: Calculator, Definition, Examples,” 2024. churnkey.co
- Finsi.ai, ecommerce repeat-purchase and category retention benchmarks. finsi.ai
- RJMetrics, “Ecommerce Buyer Behavior” benchmark, 2015 (176 retailers, 18M customers; 32% chance of a second order, 54% of a third). prnewswire.com
- miniwebtool, “Retention Rate Cohort Calculator” (power-law fit, half-life, smile detector). miniwebtool.com
- Australia Post, “eCommerce Report 2026” (rising online purchase frequency; nearly half of Millennials buying online weekly or more). auspost.com.au