Returns are usually discussed as a rate and experienced as a delay, and the delay is the part that breaks monthly reviews. This study takes 9.357 refunds from an example store's fourteen-month order book, measures how long each one took to arrive after its order, and shows what that distribution does to any figure computed before the returns have finished coming in. The median refund lands 18 days after the order it reverses, the mean is 18,5 days, and only about a fifth arrive inside the first week. Everything uncomfortable about reading a young month follows from that shape.
- Median lag 18 days, mean 18,5: most refunds arrive in the second half of the month after the order.
- Only 21,7% arrive within a week; 37,7% land in the third and fourth week; 20,4% in the second month.
- So a month read at its own close is missing most of the refunds its orders will eventually produce.
- This is exactly why refunds are booked on their issue day: reaching backwards would make every past month drift.
The lag curve
The distribution is not a spike, it is a long shoulder. About a fifth of refunds arrive in the first week, another fifth in the second, and the largest group, over a third, lands in the third and fourth week. A further fifth arrives in the second month, and a small tail comes later still.
That shape is the product of ordinary human behaviour rather than of policy: the customer waits, decides, finds the packaging, posts the parcel, and the warehouse processes it. No single step is slow and the sum is three weeks.
| Arrived | Refunds | Share |
|---|---|---|
| Within a week | 2.029 | 21,7% |
| Second week | 1.812 | 19,4% |
| Third and fourth week | 3.526 | 37,7% |
| Second month | 1.909 | 20,4% |
| Later than two months | 81 | 0,9% |
Why a young month always flatters itself
Read a month on the first of the next one and roughly 40% of the refunds its orders will produce have not been issued yet. The revenue is complete and the reversals are not, so the margin is overstated by whatever those refunds will eventually take.
It follows that comparing a fresh month against a matured one is always unfair to the older month, and that a return rate quoted without its maturity will change. The honest comparison is at matched maturity: the same number of weeks after each period, or nothing.
Why the refunds still belong to the day they were issued
The alternative is to push each refund back onto the month of its original order, which makes any single month internally tidier and makes the whole history unstable: every past month improves as the present one bleeds, so no period ever settles.
Booking each refund on its issue day is noisier per month and correct across the series, which is the trade explored in which day a refund belongs to. This study is the empirical case for it: with a median lag of 18 days, backdating means every month keeps moving for two months after it ends.
What the delay does to cash
Refunds drain the month after the one that earned them. A strong November sends its returns into December and January, which is why a busy season is so often followed by a month that feels inexplicably tight despite the previous one having been excellent.
The practical response is a reserve sized on the trailing return rate against your strongest recent month rather than your average one, because the reserve exists for exactly the month that follows a spike.
The cost, not the refund
None of the above prices a return; it only times it. The damage on an ordinary order is the margin handed back plus the outbound parcel, the uncredited payment fee, the return label, the bench time and any markdown on goods that cannot go out again, which is reliably larger than the refund suggests and which a return cost calculator will work out on your own figures.
The two findings compound: returns cost more than they look like and they arrive later than they feel like. A store that prices them properly and reads them at matched maturity is doing the only two things that make a return rate actionable.
What a monthly review should therefore look like
Read the month that closed six weeks ago rather than the one that closed yesterday, or read yesterday's with an explicit note that its returns are about sixty percent complete. Both are honest; only one of them is usually what happens.
For a trend, compare each month at the same age: the figure as it stood four weeks after close, every time. It is a small discipline and it removes an entire category of false conclusions, including the seasonal favourite where a strong December appears to have unusually good returns behaviour until January finishes arriving.
About this data
9.357 refunds from an example store's own order book, generated rather than taken from any customer, measured in whole days between the order timestamp and the refund timestamp. The lag distribution is a property of that store's simulated behaviour, so treat the shape as illustrative and your own numbers as authoritative; the method behind every figure on this site is described in how every figure is computed.