Return Rate: The Number That Decides Whether Your Ecommerce Works

A return is a sale that cost you money

The accounting instinct is to treat a return as an erased order: revenue reverses, cost of goods reverses, and the ledger returns to where it started. It does not. Four costs stay behind. You paid to ship the parcel out. You paid, or the customer paid, to ship it back. Somebody opened it, checked it, repackaged it and put it away. And the unit itself is no longer identical to new stock — it may go straight back on the shelf, it may need a markdown, or it may be unsellable. Only the first of those four is sometimes zero.

Put a $100 order under a microscope with $6 of outbound shipping, $6 of return shipping and $4 of handling and inspection. If the unit goes back into full-price stock, the returned order leaves a $16 hole. If it has to be sold 20% off, add another $20 and the hole is $36. If it cannot be resold at all and the goods cost you 60% of the price, the hole is $16 + $60 = $76. Not zero. Not a wash. A negative contribution that some other order has to cover.

That last sentence is the whole article. Because returns have negative contribution, they are not funded by their own revenue — there is none — but out of the margin of the orders that stayed sold. The moment that transfer exceeds the margin available, the whole cohort of orders is loss-making, no matter how good the conversion rate on the front end looked.

The break-even return rate, and where it bites

Write it out. Out of N orders, a proportion r comes back. The orders that stayed sold contribute N(1−r) × m × P, where m is the contribution margin and P the order value. The returned ones consume N × r × L, where L is the total loss on a returned order. Set the two equal and N cancels, leaving r* = mP ÷ (mP + L). No forecast, no assumption about growth, just the point at which the two sides of the ledger meet.

The table above runs that formula across five margins and three product profiles. Two things jump out. The first is how much the fate of the returned unit matters compared with the margin: at a 40% margin the break-even rate is 71.4% if the unit goes back to full price, 52.6% if it needs a 20% markdown, and 34.5% if it is written off. The product's second life is worth more than twenty points of tolerance.

The second is a small piece of algebra worth keeping. When the unit is written off, L = $16 + (1 − m) × $100, so mP + L = 100m + 16 + 100 − 100m = $116 regardless of the margin. The break-even rate collapses to the margin multiplied by 100 ÷ 116 — a straight line. At a 20% margin it is 17.2%, at 40% it is 34.5%, at 60% it is 51.7%. Any category where returned goods are effectively scrap has a break-even rate slightly below its own margin, which is a brutally simple rule of thumb: if you cannot resell what comes back, you cannot afford a return rate anywhere near your gross margin.

Who pays for what, in the EU and in the United States

The two markets are genuinely different here, and generalising is how merchants get the model wrong. In the European Union, Directive 2011/83/EU gives a consumer buying at a distance a right of withdrawal of 14 days, without having to give a reason. That is a statutory floor: a shop can offer 30 or 100 days, but it cannot offer 7. The trader must reimburse the payment received, including the cost of standard delivery, so the outbound shipping in the ledger above is a loss the merchant cannot contract away. The consumer bears the direct cost of returning the goods only if the trader informed them of that before the order — otherwise the trader pays that too. And if the customer handled the goods beyond what is needed to establish their nature, characteristics and functioning, the trader may hold them liable for the diminished value. A defined list of exceptions applies, including personalised goods, perishables and sealed audio, video or software once unsealed.

The United States has no federal equivalent for an online purchase. The Federal Trade Commission's Cooling-Off Rule does give three business days to cancel, but it applies to sales made at the buyer's home, workplace or a temporary location such as a hotel room or a fairground — not to purchases made on a website. Federal law does regulate how fast you must ship or refund an order you cannot fulfil, under the Mail, Internet, or Telephone Order Merchandise Rule, but it does not create a right to change your mind. In practice a US returns policy is whatever the merchant publishes, subject to state-level disclosure rules, which is why the American market shows far more variation in window length, restocking fees and return shipping than the European one.

The practical consequence for the model is that the same product can carry two different values of L in two markets. A European merchant who informs customers correctly can push the $6 of return shipping onto the buyer, but never the $6 outbound. A US merchant can, in principle, push both — and most choose not to, because free returns are a conversion lever. Which brings us to the decision nobody can avoid.

Bracketing, and why your order count stops meaning anything

Some categories are structurally high not because their customers are difficult but because the product has an attribute you cannot verify before delivery. Fit is the obvious one: nobody can tell from a photograph whether a shoe will be a 42 in this particular maker's cut. Colour under domestic lighting, fabric weight in the hand and the exact shade of a paint sample are the same problem. In those categories the buyer resolves the uncertainty by ordering the whole range and sending back the rest — the behaviour usually called bracketing.

Model it explicitly. A customer intends to keep one item and orders three sizes in one parcel. Two come back. The item-level return rate for that customer is 2 ÷ 3 = 66.7%, and the keeper rate is 33.3%. Now look at the order count: it is one order, and depending on how your system classifies a partial return it may register as an order with a return, an order without one, or a third of a return. The order count has stopped carrying information, because one order no longer corresponds to one purchase decision.

The economics are less alarming than the percentage. At $100 a unit and a 50% contribution margin, the kept item contributes $50. The parcel went out once ($6), came back once ($6), and two units were inspected and restocked ($4 each, so $8). Net contribution: $50 − $6 − $6 − $8 = $30, against $44 for a clean single-unit order that stayed sold. Bracketing costs $14 per order and still leaves 68.2% of a clean order's contribution — provided the returned units go back into full-price stock. If they need a markdown, the same order turns negative. That condition, not the headline return rate, is what you should be managing.

So measure the keeper count, not the order count. Units kept divided by units shipped, per SKU and per size variant, is the only number that survives bracketing intact — and it points at the fix, because a size that is kept far less often than its neighbours is a sizing chart problem, not a customer problem. For context on the scale of the phenomenon: the National Retail Federation, with Happy Returns, put United States retail returns at $849.9 billion in 2025, 15.8% of annual sales, with online sales returned at 19.3% and 9% of all returns judged fraudulent — down from 16.9% in 2024.

A stricter policy is a bet on conversion

Here is the decision, modelled rather than argued. Start with 100,000 sessions, a 2.50% conversion rate, a 30% return rate, a 50% contribution margin on a $100 order and a $36 loss per returned order. Expected contribution per order is 0.70 × $50 − 0.30 × $36 = $24.20. Multiply by 2,500 orders and the cohort contributes $60,500.

Now tighten: charge a $6 return fee and shorten the window to whatever the law allows you. Suppose that cuts the return rate to 22% and recovers $6 of the $36 loss, so the loss per return is $30. Expected contribution per order becomes 0.78 × $50 − 0.22 × $30 = $32.40, a 33.9% improvement per order. The question is what it costs on the way in. Set the new cohort equal to $60,500 and you need $60,500 ÷ $32.40 = 1,867.28 orders, which is a 1.867% conversion rate. The policy therefore survives a fall in conversion of up to 25.31%, and loses money beyond it.

Both outcomes are inside the plausible range, which is why this cannot be decided by opinion. If conversion holds at 2.20% the cohort contributes 2,200 × $32.40 = $71,280, up 17.8%: an excellent decision. If conversion falls to 1.70% the cohort contributes 1,700 × $32.40 = $55,080, down 9.0%: the merchant cut returns by more than a quarter and destroyed $5,420 of contribution doing it. Same policy, same return improvement, opposite verdicts, and the deciding variable sits on the other side of the funnel entirely.

So run it as an experiment, not a decree. Change the policy on a randomised share of traffic or in one market, and measure conversion and returns on the same order cohort — which means waiting out the full return window before reading the result, because returns lag sales and an early read will always flatter the stricter arm. The sibling article in this series on cart abandonment covers the trap on the other side of the same funnel: an intervention that lifts completed carts is worth nothing if the extra orders come back.

Frequently asked questions

How do I calculate a return rate correctly?

Divide units returned by units shipped from the same cohort of orders, not by units shipped in the same calendar month. Returns lag sales by the length of the return window, so a merchant whose sales grew 20% this month will divide this month's returns by an inflated denominator and understate the rate. Fix the denominator to a cohort — for instance every order placed in March — and read the rate only once March's return window has closed. Report units rather than orders wherever customers order multiple variants at once.

Does a return cost me the full value of the order?

No — it costs you the margin you would have earned plus the cash the round trip consumed. On a $100 order at a 50% margin, a kept order contributes $50 while a returned one costs $16 if the unit goes straight back to full-price stock, $36 if it needs a 20% markdown, and $66 if it is written off. That is why the break-even return rate is a range and not a constant, and why two shops with identical return rates can have opposite profitability: the difference is entirely in what happens to the unit after it comes back.

Can I charge customers for return shipping?

In the European Union, yes for the direct cost of sending the goods back, but only if you informed the consumer of it before the order was placed; if you did not, you bear it. You must still refund the payment including the standard delivery cost, and you cannot offer a withdrawal window shorter than the statutory 14 days for a distance sale. In the United States there is no federal right of withdrawal for an online purchase at all, so the terms are those you publish, subject to state disclosure rules. Model the two markets separately — the same product genuinely carries two different costs per return.

What counts as a normal return rate?

The one published figure worth quoting is the National Retail Federation's, produced with Happy Returns: United States retail returns of $849.9 billion in 2025, equal to 15.8% of annual sales, with online sales returned at 19.3% and 9% of all returns judged fraudulent — against 16.9% in 2024. Below that headline the distribution is extremely wide and skewed by category: goods whose fit or exact appearance cannot be verified before delivery sit far above it, while consumables and goods with a single specification sit far below. Compare yourself with your own category and your own break-even rate, never with a retail-wide average.

Should I block customers who return most of what they buy?

Measure before you judge. A customer who orders three sizes and keeps one shows a 66.7% item return rate, and on the model above still leaves $30 of contribution on a $100 unit at a 50% margin — 68.2% of what a clean single-unit order leaves. That customer is profitable, just less so. The ones worth acting on are those whose returned units cannot go back to full-price stock, and the right first response is a sizing chart, better photographs or a fit guide, not an account restriction. Reserve enforcement for genuine returns fraud, which the National Retail Federation puts at 9% of returns.

How does the return rate interact with conversion?

They are not independent, which is why optimising either one alone is unsafe. A generous returns policy raises conversion and raises returns; a strict one lowers both. The only figure that settles the question is the expected contribution per order, which combines them: (1 − r) × margin − r × loss per return. On the model in this article that is $24.20 with free returns at a 30% return rate and $32.40 with a $6 fee at a 22% rate, so the stricter policy pays as long as conversion falls by less than 25.31%. Run the change as an experiment on a randomised share of traffic and read the result only after the return window has closed.