Ecommerce Return Fraud Review Statistics 2026: Claim Rates, Review Time, and Loss Data

Ecommerce return fraud review statistics can look more precise than the underlying evidence. One study may report the share of return value that retailers classify as fraud. Another may combine merchandise returns with claims and appeasements. Neither figure is the same as a merchant's confirmed net loss after investigation and recovery.

This review keeps the denominators visible. It separates all returns, suspected or estimated fraud, review decisions, and realized loss. That distinction matters because a risk score is a reason to inspect a claim, not proof that the customer committed fraud.

Ecommerce return fraud statistics at a glance

Sources: NRF and Happy Returns, 2025, Appriss Retail, 2024, US Census Bureau, second quarter 2026, and Happy Returns, 2025.

The two fraud percentages should not be presented as a trend. They come from different studies, years, scopes, and methods. The 2024 Appriss measure combines returns and claims fraud and abuse. The 2025 NRF figure describes all returns estimated as fraudulent. A change from 15.14% to 9% does not prove that fraud fell.

Total returns are the first denominator

NRF and Happy Returns projected that US retailers would receive $849.9 billion in merchandise returns in 2025, equal to 15.8% of annual sales. Their online estimate was higher at 19.3%. The research included 358 ecommerce professionals at large US merchants and 2,006 consumers who had made an online return during the previous 12 months.

Those figures measure sales value. They do not say that 19.3% of online orders, packages, items, or customers generated a return. A high-value return affects a sales-value rate more than several low-value items.

The Census Bureau supplies useful scale but not a return rate. It reported $340.2 billion in seasonally adjusted US ecommerce sales for the second quarter of 2026, or 17.1% of $1.9865 trillion in total retail sales. Census publishes sales estimates and does not identify which sales were later returned or fraudulent.

For operating context beyond fraud review, see ecommerce return processing statistics and ecommerce returns customer support workload statistics.

Suspected fraud is not confirmed loss

A defensible dashboard needs at least four stages:

1. A return or claim enters the system.
2. Rules or a risk model flag it for review.
3. An authorized reviewer approves, denies, or escalates it.
4. Finance records the actual loss, recovery, chargeback, or inventory outcome.

The counts usually shrink as a case moves through those stages. Treating every flag as fraud overstates the problem and hides false positives. Treating every denied refund as recovered cash also overstates prevention because the merchant may face an appeal, chargeback, replacement shipment, or customer-service cost.

NRF's 2025 research estimated that 9% of returns were fraudulent. Retailers that tracked specific schemes reported increases in overstated quantities, empty-box or "box of rocks" returns, and decoy merchandise. These are survey results about retailer experience. They are not a transaction-level public dataset that independently verifies each case.

The distinction also appears in academic work. A 2023 qualitative study of fraudulent returns in multichannel retailing used interviews with retail experts to map how policy, process gaps, and channel differences enable abuse. The authors describe return fraud as difficult to measure and note that retailer-side evidence has been limited. The peer-reviewed study supports careful classification, not a universal fraud rate.

What the $103 billion loss estimate covers

Appriss Retail estimated that returns and claims fraud and abuse cost US retailers $103 billion in 2024. It placed total merchandise returns at $685 billion, or 13.21% of $5.19 trillion in retail sales, and classified 15.14% of returns as fraudulent.

The study combined data from more than 60 large US retailers with Census data and surveys of 150 retail executives and 1,000 consumers. Appriss said its network covered one-third of US omnichannel sales and 150,000 retail locations. That is substantial coverage, but the $103 billion figure remains an industry estimate. It is not a sum of audited, confirmed losses reported by every retailer.

Scope matters here. "Claims" can include a customer saying an order never arrived, an item was damaged, or a box was empty. A merchandise return is a different workflow because a physical item may come back for inspection. Combining them gives a broader view of exposure, but it cannot tell an ecommerce manager the loss rate for returns alone.

Retailers should report three dollar measures:

The third measure is the closest to realized loss. It should include a documented outcome and accounting period.

Public research does not give a universal review-time benchmark

The reviewed NRF, Happy Returns, Appriss, Census, and academic sources do not publish a cross-retailer median for fraud-review time. There is no authoritative public figure such as "the average return claim takes four hours" that applies across ecommerce categories.

That gap is understandable. Some merchants approve low-risk returns at initiation. Others refund at the first carrier scan, at drop-off after item verification, after warehouse receipt, or only after inspection. A review may cover identity, order history, tracking, item condition, serial numbers, photos, or an appeal. These workflows do not share one start or finish point.

Consumer research does show why timing matters. In the 2025 NRF and Happy Returns study, 84% of shoppers said box-free, label-free returns with instant refunds were their favorite return method. NRF later described a risk-based approach that preserves immediate refunds for most shoppers while routing a smaller flagged group to extra review. That design requires separate timing measures for straight-through and reviewed claims.

Measure review time with recorded timestamps:

queue time = review started at - claim submitted at

handling time = decision recorded at - review started at

customer wait time = refund or final notice at - claim submitted at

Report the median and 90th percentile. The average alone can hide a small backlog of old cases. Segment the results by claim type, risk band, decision, value, sales channel, and whether physical inspection was required.

A claim-review funnel that can be audited

Consider a planning example. It is not an industry benchmark. An ecommerce business receives 10,000 return or delivery claims in a month. Its screening system routes 700 to review. Reviewers confirm policy abuse in 210 cases and deny $31,500 in requested refunds. Later appeals reverse $3,000, while investigation and replacement costs add $1,500.

• Flag rate: 700 / 10,000 = 7.0%
• Confirmation rate among reviewed claims: 210 / 700 = 30.0%
• Confirmed rate across all claims: 210 / 10,000 = 2.1%
• Gross value denied: $31,500
• Net prevented loss before labor cost: $31,500 - $3,000 - $1,500 = $27,000

The business should not publish a 7% fraud rate. Seven percent was the screening rate. The 2.1% confirmed rate is better supported, provided each of the 210 decisions used a documented standard. Even that number describes the merchant's own claims during one month, not the retail industry.

Common return and claim fraud patterns

The 2024 Appriss study reported that 60% of surveyed retailers encountered wardrobing, 55% saw returns involving fraudulent or stolen tender, and 48% faced stolen-merchandise returns. These percentages are shares of surveyed retailers reporting each pattern, not shares of all return transactions.

NRF's 2025 discussion of ecommerce return fraud also identifies empty packages, label tampering, price switching, overstated quantities, and no-proof returns. The point at which a refund is released changes the exposure. A carrier scan proves that a parcel entered a network. It does not prove that the correct product and quantity are inside.

Review procedures should specify the evidence required for each pattern. Serial-number mismatches, item-weight differences, delivery scans, account history, and inspection photos can support a decision. None should be treated as conclusive outside its context.

Metrics for a return-fraud review dashboard

A useful dashboard keeps volume, speed, accuracy, and money separate:

1. Claims submitted by type and sales channel.
2. Share approved automatically and share routed to review.
3. Median and 90th-percentile queue, handling, and customer wait time.
4. Review decisions by approve, deny, escalate, and request-more-evidence outcome.
5. Appeals and reversals by original decision reason.
6. Confirmed fraud rate across all claims and among reviewed claims.
7. Gross value denied, recovered value, confirmed loss, and review labor cost.
8. False-positive proxy, such as appealed denials later reversed.
9. Customer contacts and repeat contacts per reviewed claim.
10. Reviewer agreement and quality-audit results.

Compare like periods and keep policy changes visible. A stricter threshold may lower losses while raising review volume, customer wait, appeals, and support contacts. Those effects belong in the same evaluation.

Where trained support fits

An ecommerce virtual assistant can collect order records, check tracking events, organize customer evidence, maintain reason codes, and prepare routine cases for an authorized reviewer. The services overview covers broader support options.

Final fraud decisions, high-value refund holds, account restrictions, and write-offs should stay with staff who have defined authority and access controls. Give support personnel only the data needed for their assigned step. Log every access and change, and keep an escalation path for cases that do not fit the procedure.

Frequently asked questions

What percentage of ecommerce returns are fraudulent?

No universal ecommerce-only confirmed rate exists in the reviewed public sources. NRF and Happy Returns estimated that 9% of all US retail returns were fraudulent in 2025. The figure covers retail returns and should not be applied to one merchant without checking channel, product mix, and method.

How much does return fraud cost retailers?

Appriss Retail estimated $103 billion in US return and claims fraud and abuse for 2024. The study used retailer data, Census data, and executive and consumer surveys. It is an industry estimate with a broader scope than confirmed ecommerce merchandise-return loss.

How long should a return-fraud review take?

Public industry research does not establish one median review time. Set separate service levels for queue time, analyst handling time, and total customer wait. Report the median and 90th percentile for reviewed claims, and keep automatic approvals out of the analyst-time measure.

Is every flagged return fraudulent?

No. A flag indicates that a rule or model found a reason for review. A confirmed case needs evidence, a recorded decision, and a consistent definition. Track flags, decisions, appeals, and financial outcomes as separate events.

Sources and methodology

This review uses the 2025 NRF and Happy Returns retailer and consumer research, the 2024 Appriss Retail return and claims study, the second-quarter 2026 US Census ecommerce release, NRF's 2025 and 2026 return-fraud discussions, and peer-reviewed research on fraudulent returns and return-policy timing.

The article does not average incompatible fraud estimates. It labels the year, sample, channel, and denominator for each material figure. The worked funnel is arithmetic from stated assumptions, not observed industry performance. The 2026 title identifies this review edition; it does not imply that every source was collected in 2026.