How do I forecast second, third and fourth purchases from customer cohorts?
By Trayan Hristov · Founder, Thrivelia · 19 November 2026
If 25% of first-time buyers return, how many fourth orders can you count on? You can’t answer from the first repeat rate alone. The people who reach order two are a selected group, and they start a new clock. I’d show the founder three transition rates, the time each takes and the contribution attached to each order.
Start a new clock after every qualifying purchase
Build one row per customer and paid order. Sort by paid time using a stable order ID for ties. Collapse split shipments and remove replacements, test orders and duplicate transactions. Decide upfront how cancelled or fully refunded orders affect eligibility and contribution. For this example a qualifying purchase is paid and kept after the return window. Let the return window mature before closing the cohort. In a subscription business, show automatic renewals separately from purchases the customer actively chose.
For a 90-day first-to-second rate, the denominator is buyers observed for 90 days after order one. The numerator is those who placed order two within those 90 days. For second-to-third, require a full 90 days after order two. Apply the same rule after order three. Use a single acquisition cohort with sufficient follow-up for the whole sequence, or estimate each conditional rate from comparable stage cohorts with the same entry products, seasons and business rules.
Transition in a fictional cohort | Eligible | Advanced | Conditional rate |
Order one to two within 90 days | 1,000 | 250 | 25% |
Order two to three within the next 90 days | 250 | 100 | 40% |
Order three to four within the next 90 days | 100 | 50 | 50% |
Translate the rates into money and time
For 1,000 new buyers with that mix, the scenario produces 250 second orders, 100 third orders and 50 fourth orders. That’s 400 repeat orders. At £28, £32 and £36 contribution per respective order, repeat contribution is £7,000 + £3,200 + £1,800 = £12,000. Deduct incentives, fulfilment and variable service costs consistently. The example assumes each next step happens inside its own window. A fourth purchase could land up to 270 days after the first. This is not a £12,000 forecast for next quarter.
If the first transition improves from 25% to 27% and downstream behaviour stays the same, expected repeat contribution rises to £12,960. That leaves £960 of headroom before the costs of the intervention. A £1 voucher cost on all 1,000 entrants would already exceed it. New marginal buyers may also repeat less often than the original group. Re-estimate their downstream behaviour before scaling.
Build a forecast you can challenge next month
For a calendar forecast, group transitions by weeks since the previous order. Apply that timing distribution to each acquisition week, then feed expected second orders into the weeks when order three becomes possible. Compare early and late scenarios based on previously observed cohorts. Report the number of eligible people beside every rate. Fifty fourth purchases across a year rarely justifies a weekly forecast by size, channel and creative.
Freeze predictions before an older held-out month’s outcomes are revealed. Compare predicted and actual kept orders and contribution by stage. Measure absolute order error and signed bias. A model that repeatedly overstates order four needs fixing even if total revenue looks close. Then choose the intervention: adoption after order one, a useful next product after order two, or replenishment timing after order three. See the original first-to-second method for the order-level setup.
Run the exercise
Replace the fictional counts and contribution inputs in the Repeat orders tab. Save the dated forecast before the next cohort matures. Download the editable workbook or browse the other retention workshops.

.png)



Comments