Retail's most-cited returns statistic is that fraud cost about $103 billion in 2024. That figure was published two different ways in the same year, from bases $205 billion apart. Then it "fell" roughly a quarter in 2025 while every directional signal beside it rose. The number is not measuring behavior. It is measuring how the number was made. Here is why that matters, and what to measure instead.
The series doesn't hold together
Line up what has been published as the industry number and the incoherence is visible from across the room:
| Year | Total returns | Return rate | Fraud rate | Fraud $ | Published by |
|---|---|---|---|---|---|
| 2022 | $816B | ~16.5% | 10.4% | ~$85B | NRF / Appriss |
| 2023 | $743B | 14.5% | 13.7% | ~$101B | NRF / Appriss |
| 2024 | $890B | 16.9% | ~11.6% implied | $103B | NRF / Happy Returns |
| 2024 | $685B | 13.2% | 15.1% | $103B | Appriss / Deloitte |
| 2025 | $849.9B | ~16% | 9% | ~$76B | NRF / Happy Returns |
Look at the two 2024 rows. Two official totals for the same year, $205 billion apart, a gap larger than the entire fraud problem being measured. And both models land on exactly $103 billion of fraud through offsetting math: a larger base times a smaller rate, and a smaller base times a larger rate. The press quotes the $103 billion as one measured fact. It is two incompatible estimates that happened to collide.
There is a second problem hiding in the total-returns column: it lurches. $816 billion down to $743 billion, then two answers for 2024, then $849.9 billion, across years in which e-commerce penetration rose every single one. Returns as a share of a growing base do not move like that. Real behavior is not that jumpy. The lurching is the measurement moving, not the world.
Why the number is built to move
The headline is a composite of two instruments stitched together, and every seam leaks.
The fraud rate is mostly self-reported perception. Retailers are surveyed on what share of returns they believe is fraudulent, and there is no shared definition of fraud. One retailer counts wardrobing, bracketing, and first-party abuse; another counts only criminal empty-box and receipt fraud. Much of the fall from roughly 15% to 9% in 2025 is definitional: the publisher narrowing what counts as fraud, not consumers behaving better. To its credit, NRF says so itself. Its 2025 report notes the years are not directly comparable due to methodology changes, and it published a piece titled Rethinking return fraud in retail. When the publisher tells you not to compare years, the series is not a trend. This is worth saying plainly: nobody here is being dishonest. The category is genuinely hard to measure from outside, and the publishers flag the breaks themselves. The dishonesty enters downstream, when the market quotes the number as if it were stable.
The dollar figure is an extrapolation. A rate is applied to Census retail sales, so the headline moves whenever the assumed return rate changes, the sales denominator is revised, or the retailer sample shifts. The fraud figure then multiplies that already-shaky total by the already-shaky perception rate. Two noisy estimates, multiplied, reported to three significant figures.
And the panel is the wrong window for a DTC brand. Appriss draws on 60 of the top 100 US retailers: large omnichannel chains with heavy in-store, POS-linked returns. That is a reasonable view into Target and Best Buy. It is close to irrelevant to an apparel brand doing mail-back returns off Shopify, which is a different return channel, a different fraud surface, and a different customer. Applying a big-box fraud rate to a DTC book is a category error baked into the source.
The tell
In the same 2025 report where the fraud headline falls to 9%, the directional indicators point the other way: among retailers that track them, overstated-quantity claims are up at 71%, empty-box returns up at 65%, and decoy returns of counterfeit items up at 64%, while 93% of retailers call fraud and abuse a significant problem. A measured number that moves opposite to everything it is supposed to describe is telling you about its own method, not about the world.
Fraud and abuse are opposite problems
The word "fraud" hides two different things, and they have opposite owners.
| Hard fraud (external, criminal) | Soft abuse (first-party, legal) | |
|---|---|---|
| Looks like | Empty box, overstated quantity, counterfeit swap, refund rings | Wardrobing, extreme bracketing, serial returning |
| Nature | Episodic, prosecutable | Structural, generational, legal |
| Who owns it | Payments-security and transaction-scoring vendors | Nobody, at the customer level |
| How it gets fixed | Detect and block the transaction | Score the customer, segment the policy |
Hard fraud is a payments-security problem, and transaction-scoring vendors already own it. Soft abuse is the bigger, stickier half for a DTC brand, and it is invisible to fraud tools, because at the transaction level every event is a legitimate customer making a legitimate return. There is nothing to flag. The pattern only exists at the customer level, across orders.
The part you can actually measure
The industry number is unmeasurable from the top because a return is a customer-level behavior being reported as an aggregate statistic. Your own blended return rate commits the same failure inside your business, for the same reason: it averages a population that is wildly non-uniform. We covered the economics of that in returns are a forward-value problem; the short version is that your best customers and your worst customers can both return a lot, and the rate cannot tell them apart.
| Customer profile | Return rate | Net contribution | Right action |
|---|---|---|---|
| Loyal high-AOV buyer who returns the misfits | High | Strongly positive | Protect. Never gate. |
| Serial bracketer with a low keep rate | High | Negative | Add friction, change the policy |
| Low-touch buyer who keeps what they order | Low | Positive | Grow share of wallet |
| One-time discount buyer | Low | Marginal or negative | Do not re-acquire on paid |
Rows one and two are indistinguishable to every tool in the standard stack. Separating them is the entire game.
Pricing returns at the customer level
Four moves, in order.
1. Join the return to the customer and the original order. Start from item-level return events, not an aggregate rate, with every return tied to who made it and what they bought. This is the join fraud tools never make, because they watch one transaction at a time. It is also the only view in which abuse becomes visible.
2. Load the full reverse-logistics cost, not the refund. The refund is the small part. Add inbound shipping, inspection and processing labor, the markdown or liquidation loss on resale, and the sunk outbound shipping and payment fees. In apparel, the all-in cost of a return is commonly estimated at 20 to 40% of the item's price. Price the return at what it costs you, per customer.
3. Score net contribution, and track keep rate. Replace gross-revenue customer rankings with net contribution after returns, then track keep rate, units kept over units ordered, as the bracketing signal the rate alone misses. This is what tells a loyal high-returner apart from a parasitic one.
4. Segment the policy. Never blanket it. 72% of retailers now charge for some returns, up from 66% the year before, but blanket fees are a blunt instrument: among merchants that started charging, 47% saw complaints rise, 37% lost customers, and 34% saw average order value fall. Undifferentiated friction saves a little margin on bracketers and quietly taxes the loyal high-returners you most need to protect. Protect the customers whose netted value is strongly positive, add friction only where net contribution is negative, and stop re-acquiring the customers who never repay the cost.
The honest catch
Joining returns to customers, loading true reverse-logistics cost, and scoring net contribution forward is a real modeling problem. Doing it once in a spreadsheet is a good weekend. Doing it every week, accurately enough to move policy and budget against, is where most teams stall. That gap between the right question and the answer on a cadence is what Tacet exists to close.
The verdict
Returns fraud cannot be measured from the top, and the industry's own numbers prove it: two totals $205 billion apart in one year, a headline that falls while every signal under it rises, and a publisher that says not to compare the years. The only returns-fraud number a brand can trust is the one computed from its own item-level, customer-joined data, against its own definition of abuse.
- Never put an industry returns-fraud figure in a plan. Cite it, if at all, as illustrative, and say which of the incompatible versions you mean.
- Compute keep rate and net contribution after returns for your own customers. That number is measurable, and it is the one that moves policy.
- Before adding any blanket return fee, identify the loyal high-returners it would tax. Rows one and two of the table above are different customers deserving opposite treatment.
The free diagnostic scores every customer on 12-month forward value net of returns and discounts, on a read-only connection, and you keep the analysis either way.
Take it with you. The Returns Self-Audit is a free field guide and worksheet: the four customer profiles, the fraud-versus-abuse signals to watch, and a fill-in sheet to compute net contribution after returns on your own numbers. It runs in an afternoon, no access required.
Sources
- NRF and Happy Returns: 2024 Retail Returns to Total $890 Billion (December 2024)
- Appriss Retail and Deloitte: Fraudulent Returns and Claims Cost Retailers $103B in 2024 (December 2024)
- NRF and Happy Returns: 2025 Retail Returns Landscape and the companion release (October 2025)
- NRF: Rethinking return fraud in retail (May 2026)
- Retail Dive: Retailers took back $743B in returns last year, with $101B lost to fraud (December 2023, covering the 2022 and 2023 NRF and Appriss figures)
- DC Velocity: Over 15% of all retail returns in 2024 were fraudulent (January 2025)