NRF and Happy Returns estimated that 15.8% of total retail sales and 19.3% of online sales would be returned in 2025. The two figures describe different sales-channel denominators. Use the online figure when discussing the report's ecommerce benchmark; use the overall figure when discussing its combined retail estimate. Neither is a universal target for an individual store. [1]
The two headline rates have different denominators
A return rate is a fraction. Its meaning depends on both the returned amount above the line and the eligible sales below it. Switching the denominator changes the question even when the label stays the same.
NRF's release projects $849.9 billion in total retail returns for 2025. Its overall annual-sales estimate includes a broader retail population than its online-sales estimate. The release was published on October 15, 2025; those figures should retain their forecast wording and year when cited. [1]
| Reported measure | 2025 estimate | Appropriate description | Avoid this interpretation |
|---|---|---|---|
| Total annual retail sales returned | 15.8% | Overall retail estimate across channels | The return rate of every ecommerce store |
| Online sales returned | 19.3% | Online-sales return estimate | The share of online shoppers who returned anything |
| Total retail returns value | $849.9 billion | Projected total industry return value | The total cost of processing returns |
Source for the table: NRF's 2025 announcement. Returned merchandise value and processing expense are different quantities. Shipping, inspection, repacking and inventory write-downs require their own accounting.
Returned revenue, returned orders and returned items are separate metrics
A merchant can calculate several valid return rates from the same order file. Problems arise when the dashboard calls each one simply “returns.”
Consider an illustrative store with 100 orders, 150 units sold and $10,000 in merchandise sales. Suppose 10 orders generate returns involving 15 units worth $1,500. The order return rate is 10%, the unit return rate is 10%, and the returned-value rate is 15%. These are hypothetical calculations, not observations from NRF or TrueMargin customers.
The value rate is higher because the returned units in this example are more expensive than the average unit sold. A retailer could see the opposite pattern when shoppers keep higher-priced merchandise and return inexpensive add-ons. An order with several line items also remains one order when a single item comes back.
Record all of these measures if they help operations, but give them explicit names. Warehouse workload is connected to physical units and parcels. Revenue reconciliation uses money. Customer-service analysis may focus on affected orders or customers. A return-rate benchmark only becomes useful after its measurement matches the decision.
Return timing can make a healthy sales week look worse
Returns processed during a week usually do not describe only the purchases made during that week. A processing-date report is useful for cash movement and operations, but it can produce a distorted product-quality signal.
Imagine a hypothetical business that sells $40,000 in a promotional week and $10,000 the following week. If $3,000 from the promotion is returned during that quieter week, dividing those returns by the quieter week's sales produces 30%. That number describes current-period activity; it does not establish that 30% of the quieter week's purchases came back.
For product diagnosis, group purchases by their original order date and follow their returns forward. Keep the completed observation window clearly visible. A cohort still inside its return window is incomplete, and the newest cohort may appear artificially strong simply because customers have had less time to send items back.
Maintain a second view for actual refunds processed by date. That view supports cash planning. The cohort view supports comparison. Neither should silently replace the other, and refund processing should not be assumed to coincide with the warehouse's receipt of the merchandise.
A blended store benchmark can hide a category problem
A combined return rate can change when your sales mix changes, even if the return behavior within each product group stays steady. That is an arithmetic effect worth checking before attributing a movement to a policy or creative change.
For a hypothetical example, give category A a 10% returned-value rate and category B a 30% rate. Equal sales in both produce a 20% blended rate. If category A later represents 75% of sales and category B represents 25%, the blended rate becomes 15%, without either category improving.
The calculations use sales-value weights: 0.5 × 10% + 0.5 × 30% = 20%, followed by 0.75 × 10% + 0.25 × 30% = 15%. These are explanatory scenarios, not market benchmarks.
Split your analysis by the dimensions you can act on: SKU family, size, acquisition channel, geography and return reason. Do not fragment the data so far that one unusual order dominates every comparison. Start with the largest returned-value contributors and inspect the orders behind the percentage.
Return rate is an input to profitability, not a complete cost measure
Returned merchandise does not always become a total economic loss. A resalable item may return to stock, while a damaged item may need a write-down. An exchange also has a different revenue outcome from a full refund.
A practical return-cost worksheet separates refunded revenue, recoverable inventory value and incremental handling expenses. Include the shipping and service costs the merchant actually bears. Check whether fees are retained by providers rather than assuming that every original transaction cost reverses with the refund.
Then connect the return to its original acquisition cohort. Advertising can look successful on booked sales while becoming less attractive after refunds settle. Conversely, treating every returned item as worthless can overstate the damage when inventory is recovered and later sold.
At TrueMargin, the useful analytical distinction is between money collected, revenue retained and contribution remaining after variable costs. A national returned-value estimate is context for that work. It cannot replace your actual refund ledger, inventory recoveries or fulfillment bills.
A source card makes the benchmark safer to reuse
Store the research context alongside the number instead of asking each writer to reconstruct it. A citation card should identify the publisher, report year, sales channel, unit, geography, estimate status and original URL.
NRF describes two summer 2025 surveys: 2,006 consumers who had returned an online purchase during the previous year, and 358 ecommerce professionals at large U.S. merchants with more than $500 million in revenue. That is important population context, not a claim that the findings precisely represent a small Shopify brand. [1]
Do not merge the consumer survey's attitude questions with the merchant survey's sales estimates. A person saying that free returns matter is answering a different question from a merchant estimating returned merchandise value.
Also preserve the report's status as an estimate made during the year. An article published later may analyze that estimate, but its new publication date does not turn the underlying evidence into a new annual study.
Methodology and limitations of this analysis
This article reconciles the definitions in NRF's public 2025 announcement and adds clearly labeled arithmetic examples. TrueMargin did not collect the NRF surveys or independently audit merchant responses. The scenarios explain denominator and timing effects; they do not estimate a reader's expected results.
The public release does not provide enough detail to reconstruct every weighting or channel calculation. We therefore do not derive an implied in-store rate, calculate a confidence interval, or claim a statistically significant difference between channels. Those questions require the underlying methodology and appropriately matched data.
When using the figures for planning, pair them with a range of your own cohort outcomes. A benchmark is most valuable when it prompts a better question about your store, rather than becoming an unsupported pass-or-fail target.
Frequently asked questions
Which NRF return rate should an ecommerce article cite'
Use the online-sales estimate when describing NRF's ecommerce figure, and retain the report year and estimate wording. Do not replace it with the overall retail rate merely because that figure appears in the headline.
Does the online rate measure the share of shoppers making returns'
No. A sales-based return rate and a shopper participation measure have different denominators. Keep the source's measurement attached to the number.
Can I calculate an in-store rate from the two headline figures'
Not from those figures alone. You would need compatible sales-channel weights and definitions, and should not invent the missing inputs.
Is a refund rate the same as a return rate'
It depends on the reporting definition. A refund may occur without a physical return, while an exchange may involve returned merchandise without a full revenue reversal. State exactly what your report counts.
Should a new analysis use the current year in every statistic'
No. The article's analysis date and the source's data year should remain separate. Cite older evidence honestly until a newer comparable study is available.
Related research: see a comparison of retailer return fees for a worked application of these measurement distinctions.
Sources
- NRF and Happy Returns: Consumers Expected to Return Nearly $850 Billion in Merchandise in 2025, October 15, 2025. Primary source for the reported estimates and survey populations.