How to Calculate Return Rate: 5 Ways the Number Lies

How to calculate return rate is a two-minute question with a five-part answer, and most sellers get at least two parts wrong. The formula is trivial — returns divided by sales. The trouble is that "returns," "sales," and "divided by" each hide a decision, and the wrong decision produces a number that looks precise, gets reported to a board, and tells nobody what to fix. These are the five failures, in the order they cost you money, with the fix for each.

How to calculate return rate: the formula and its three hidden decisions

Return rate is the share of what you sold that came back. There are two legitimate versions:

Unit return rate = units returned ÷ units sold, over the same order cohort.

Value return rate = returned revenue ÷ gross revenue, over the same order cohort.

The three hidden decisions are: which period the denominator covers, whether you count units or money or orders, and which returns you include at all. Get those wrong and the arithmetic is still correct — the number just stops meaning anything.

Mistake 1: matching returns to the wrong month

This is the most common and the most expensive, because it makes the rate move when nothing about your product changed.

Returns arrive after the sale, by a lag you can look up. Amazon lets buyers return most items within 30 days of delivery. In the EU, the Consumer Rights Directive gives a consumer 14 days to withdraw from a distance contract without giving a reason — and that clock starts at delivery, not at order. Add transit both ways and a perfectly ordinary return lands 40 to 50 days after the order was placed.

So dividing "returns received in March" by "orders placed in March" compares two populations that barely overlap. In a month where sales grew, the denominator inflates and the rate looks better than it is. In January, after a December peak, the denominator collapses while December's returns pour in and the rate looks catastrophic.

Fix: compute by cohort. Take the orders shipped in a period, then count the returns that came back from those orders, whenever they arrive. Report the last 30 to 60 days as provisional, because the window has not closed yet. If your reporting cannot follow an order to its return, that is the first thing to fix — everything downstream depends on it.

Mistake 2: not saying which denominator you used

Units, orders, and revenue give three different numbers for the same business, and people quote them interchangeably.

Take one illustrative month: 1,000 units shipped across 700 orders, £80,000 of gross revenue. 80 units come back, spread across 55 orders, worth £9,600.

Metric	Calculation	Result
Unit return rate	80 ÷ 1,000	8.0%
Order return rate	55 ÷ 700	7.9%
Value return rate	9,600 ÷ 80,000	12.0%

Same month, same business, and the value rate is 50% higher than the unit rate — because expensive items came back disproportionately, which is exactly what happens with furniture, lighting, and anything that has to fit a space.

Fix: pick one as the headline metric and label it every time. Unit rate tracks customer behaviour and is the right operational number. Value rate is what finance needs for the returns reserve. Order rate matters only if your fulfilment cost is per-order. Publish all three internally; never let them get quoted without their label.

Mistake 3: including returns you could never have prevented, then trying to act on the total

Not every return is a signal about your listing. The National Retail Federation's 2025 returns research found that 9% of all returns are fraudulent. Add carrier damage, wrong-item picks from a 3PL, and cancellations that arrive as returns, and a meaningful share of the total has nothing to do with how you described the product.

Fix: keep two rates.

Gross return rate — everything that came back. This is the finance number and the marketplace-facing number.
Preventable return rate — gross minus fraud, carrier damage, fulfilment errors, and cancellations. This is the number your listing work moves.

If your gross rate is 14% and your preventable rate is 9%, then the ceiling on what better images and better specs can achieve is 9 points, not 14. Setting a target against the wrong one guarantees you miss it.

Mistake 4: benchmarking a blended rate against someone else's blended rate

Published benchmarks are useful and routinely misused. The NRF forecast a 15.8% overall retail return rate for 2025, worth $849.9 billion, down from 16.9% and $890 billion in 2024. But that figure covers all of retail including physical stores. For online sales specifically, the same research put the expected 2025 rate at 19.3%, and retailers forecast 17% of holiday sales coming back.

A furniture seller whose rate is 12% is not "beating the industry" by comparing against 15.8%. They are comparing a single category, sold online, against a blend that includes grocery and in-store purchases.

Fix: compare against your own category, your own channel, and your own past. Category context is broken down in size return rate by category, and the aggregate figures on size-driven returns are collected on the ecommerce returns size statistics page. Your own trailing twelve months, cohort-matched, is a better benchmark than any published average.

Mistake 5: a rate with no reason breakdown

This is the one that makes the whole exercise pointless. A single percentage is a thermometer. It tells you that something is wrong; it cannot tell you what.

And the reason data is messier than it looks, because the reason attached to a return is a choice the buyer made from a dropdown, not a diagnosis. eBay's own returns process separates "remorse" returns — the buyer changed their mind — from returns where the item arrived damaged or did not match the description, and handles them differently. A buyer whose chair does not fit the space may pick "no longer needed," "doesn't fit," or "not as described" depending on which they think gets them a free label. One physical cause, three codes.

Fix: collapse the marketplace codes into cause families before you analyse anything, then act on the family:

Cause family	Codes it usually hides behind	What actually fixes it
Size or fit wrong	doesn't fit, not as described, no longer needed	Measured dimensions on the main image; clearance and assembled-size callouts
Expectation gap (colour, finish, material)	not as described, quality not adequate	Colour-accurate photography; material and finish stated in figures, not adjectives
Arrived damaged	arrived damaged, defective	Packaging spec, carton drop testing, carrier mix
Fulfilment error	wrong item sent, missing parts	Pick accuracy, kit completeness, packing list
Genuine remorse	changed my mind, found better price	Little you can do; track it so it stops polluting the other four
Fraud / abuse	any of the above	Policy and enforcement, not listing work

Once the returns are in families, the rate becomes actionable. A 9% preventable rate that is 60% size-and-fit is a completely different project from a 9% preventable rate that is 60% transit damage.

What counts as a good return rate

There is no universal answer, and anyone who gives you one is selling something. What you can say precisely:

Compare against your channel: roughly 19.3% expected for online retail overall in 2025 per NRF, versus 15.8% for retail as a whole.
Compare against your category, not the blend.
Compare against your own cohort-matched trailing trend — the only benchmark that controls for your product mix, your price point, and your customers.
Judge progress on the preventable rate, because that is the only part your work moves.

The single sentence worth remembering: a return rate without a stated denominator, a matched cohort, and a reason breakdown is a number you can report but cannot act on.

Next steps

In rough order of payoff:

Fix the cohort join first. Until returns can be traced back to the orders they came from, every other improvement is measured against noise.
Label your headline metric and stop letting unit, order, and value rates be quoted interchangeably.
Split gross from preventable. Set targets only against preventable.
Map codes to cause families and rank them by units, not by how loud the complaints are.
Attach money to the ranking. A percentage does not compete for budget; a cost does. The return cost calculator turns a rate into a per-return and annual figure, and the full cost stack — outbound, inbound, inspection, markdown — is broken down in hidden cost of ecommerce returns.
Then fix the biggest family. If size and fit tops the list, the intervention is not "write a better description." It is putting the real, measured dimensions onto the image the buyer actually looks at. There are several ways to do that: hire a designer per SKU, draw arrows by hand in a photo editor, or use annotation software that snaps to the measured edges of the product in the photo and exports at each marketplace's required size. What does not work is an AI image generator — it will produce a confident-looking "42 cm" for a product it never measured, which converts a size return into a not-as-described claim.
Re-measure by cohort after 60 days, not after two weeks. The window has to close before the number means anything.

Return-rate reporting checklist

Returns joined to their originating orders, not to the calendar month they arrived
Most recent 30–60 days flagged as provisional
Headline metric named explicitly (unit / order / value) everywhere it appears
Gross and preventable rates reported separately
Fraud, carrier damage, and fulfilment errors excluded from the preventable figure
Marketplace reason codes mapped to cause families before analysis
Benchmarks matched on channel and category, never a blended national average
Each cause family costed, not just counted
Targets set against the preventable rate only

FAQ

What is the formula for return rate?

Units returned divided by units sold, or returned revenue divided by gross revenue, over the same order cohort. The formula is the easy part; the accuracy lives in matching the returns to the orders they came from rather than to the month they arrived.

How do I calculate return rate when returns arrive weeks after the sale?

Use cohort accounting. Fix the denominator to the orders shipped in a period, then attribute every return back to its originating order regardless of when it lands. Treat the most recent 30 to 60 days as incomplete: Amazon's standard return window is 30 days from delivery and the EU right of withdrawal is 14 days from delivery, so late returns are normal, not anomalies.

Should return rate be based on units, orders, or revenue?

All three, reported separately. Unit rate is the operational number, value rate is the finance number, and order rate matters only when your cost is per-order. The mistake is quoting one without saying which it is — in a typical month the value rate can run 50% above the unit rate because expensive items come back more often.

What is a normal ecommerce return rate?

The NRF's 2025 research put the expected online return rate at 19.3% and the overall retail rate at 15.8%, down from 16.9% in 2024. Those are blends across all categories, so treat them as context rather than a target; the useful comparison is your own category, your own channel, and your own cohort-matched trend.

Why does my return rate not tell me what to fix?

Because a single percentage has no cause attached. Buyers pick a return reason from a dropdown, and one physical problem — a chair that does not fit the space — can be filed under "doesn't fit," "not as described," or "no longer needed." Group the codes into cause families first; only then does the rate point at a specific fix.