Wardrobing and returns abuse: how to detect it without punishing good customers
AI can score return behaviour at item and customer level, but blanket rules and opaque decisions cost loyalty. Here is how to target abuse narrowly and keep honest returners.
KEY TAKEAWAYS Summary by the editors
- Wardrobing means buying an item, using it once and returning it for a refund; it is one form of returns abuse alongside claims fraud and staff-enabled fraud.
- Appriss Retail's 2026 benchmark report estimated that $706 billion of merchandise was returned in 2025, of which $100 billion (14.2%) was preventable loss from fraud and abuse.
- Detection models work best as decision support at the returns counter or in the refund workflow, flagging patterns in transaction history rather than judging a single return in isolation.
- Appriss Retail reported that 55% of surveyed North American consumers avoided retailers with restrictive returns policies, so blanket rules carry a measurable cost.
- In the same benchmark, 80% of consumers wanted transparency about how AI makes return decisions and only 10% trusted AI outright, which argues for human review of every denial.
Wardrobing is the practice of buying an item, wearing or using it once, and returning it for a refund, and AI can help to detect it by scoring return behaviour across customers, items and channels. The risk lies in the false positives: a model that flags honest shoppers costs more in lost loyalty than the abuse it prevents. The practical answer is narrow targeting, human review and a clear policy that customers can read.
What is wardrobing and how big is returns abuse?
Appriss Retail describes wardrobing as buying something, such as a festive sweater, using it for an occasion and returning it the next day. It is one type of returns abuse. The same source also lists claims and appeasement fraud (a shopper says an order never arrived or was damaged, then keeps the item) and staff-enabled fraud, where untrained temporary staff process fake returns.
Scale estimates come mostly from vendors and trade bodies, so treat them as indicative. Appriss Retail's 2026 benchmark report put returned merchandise in 2025 at $706 billion, with $100 billion (14.2%) classed as preventable loss from fraud and abuse. Its release also described buy online, return in store as the fastest-growing abuse vector. The release contains figures that do not fully reconcile with one another (abuse is cited both as 12% of returns-related loss and as a headline dollar value), so check the full report before quoting a single number to your board.
In the UK, FashionNetwork reported in October 2025 on research by ZigZag and Retail Economics showing serial returners falling from 12% to 8% of shoppers, and 76% of the 100 largest clothing and footwear retailers tightening their returns policies. The same report noted displacement: serial returners sent back fewer items, but occasional and slow returners rose.
How does AI detect returns abuse?
Most systems combine three layers. The first is rules and limits, such as a cap on returns within a set period. The second is statistical scoring of a customer's history: return rate, time between purchase and return, mix of categories, and the gap between order value and kept value. The third is item-level checks, for example condition grading from photos at the warehouse. Appriss Retail describes AI that reads a shopper's transaction history at the returns counter and recommends whether the associate should approve or deny, with the recommendation anonymised.
| Signal | What it can show | Main false-positive risk |
|---|---|---|
| Return rate over a rolling period | Habitual over-ordering or wardrobing | Customers who order several sizes on purpose |
| Days between delivery and return | Use before returning, especially around events | Slow postal returns or travel |
| Value kept versus value ordered | Bracketing or fashion hauls | Legitimate fit exploration, particularly in jeans and tops |
| Claims of non-delivery or damage | Appeasement fraud | Carrier errors and genuine damage |
| Cross-channel pattern (buy online, return in store) | Policy arbitrage | Convenience, not intent |
| Item condition at inspection | Worn, washed or tag-less items | Subjective grading between warehouse staff |
Why do blanket restrictions backfire?
Appriss Retail's survey of North American consumers found that 55% avoided retailers with restrictive returns policies, 31% stopped shopping with a retailer after a negative return experience, and 70% spent more after a positive one. These are vendor survey results, but they match the logic of the problem: abuse is concentrated in a small group, while restrictions land on everyone. The article advises against blanket rules such as no receipt, no return, which can alienate loyal customers.
Targeted policies are the alternative. As Appriss Retail describes it, ASOS deducts £3.95 from a refund when a returner keeps less than £40 worth of items, and its Fair Use Policy is aimed at a small group of shoppers with histories of excessive returns. The point of such designs is that most customers never meet the rule.
What does a fair detection process look like?
- Define abuse in operational terms (for example, repeated returns of used items with a documented condition check), not as a high return rate alone.
- Score at the customer level over a long window, and combine the score with customer value so loyal, profitable shoppers are treated differently from rare outliers.
- Use graduated responses: a warning first, then a limit on free returns, then a refund deduction, and only then an account restriction.
- Keep a human in the loop for every denial or account action, with a short note of the reason that can be shared with the customer.
- Review outcomes monthly: appeal rates, reversed decisions, and the share of flagged customers who later returned to normal behaviour.
How should retailers handle trust, transparency and data?
Transparency matters to shoppers. In the same benchmark, 80% of consumers wanted to know how AI makes return decisions, 71% trusted human associates more than AI for approvals and only 10% trusted AI outright. A scoring model that cannot give a reason in plain language is a liability at the counter and in the call centre.
Return scoring is also profiling. Under the GDPR, personal data must be limited to what is necessary for a stated purpose (Article 5), and individuals can refuse decisions based solely on automated processing that have legal or similarly significant effects, with safeguards such as human intervention and the ability to contest the decision (Article 22). Take legal advice on whether a refund denial or account restriction counts, and document the reasoning. Accuracy applies as well: the data feeding the model must be correct, including carrier delivery scans, or you will punish customers for your own logistics errors.
How do you measure whether detection is working?
Measure net effect, not gross savings. Track abuse losses avoided, but also customer complaints, refund appeals, repeat purchase rate among flagged and unflagged customers, and conversion on pages that display your returns policy. A holdout group that is scored but not acted upon lets you estimate how many flagged customers would have behaved well anyway. Without it, a model can look effective while quietly removing profitable customers.
Finally, fix the causes that detection cannot. Poor size information and thin product content drive honest over-ordering, and a policy that treats those customers as suspects compounds the problem. Detection should be one part of a wider returns strategy, not a replacement for good fit data.
Frequently asked questions
What is wardrobing in fashion retail?
Wardrobing is buying an item, using it once, typically for an event, and returning it for a refund. Appriss Retail gives the example of a festive sweater bought for a party and returned the next day. It is one form of returns abuse, distinct from fraud involving stolen goods or false non-delivery claims.
Can AI reliably tell whether a return is abusive?
Not on a single return. Models are better at spotting patterns across a customer's history, and even then they produce false positives. Treat the output as a recommendation for a trained person, and measure appeals and reversed decisions to understand error rates.
Should retailers charge for returns to stop wardrobing?
Fees reduce some behaviour but do not target abuse specifically, and they affect honest shoppers too. FashionNetwork's October 2025 report suggests fees under £3 are changing behaviour in the UK, though it also notes displacement towards slower returns. Targeted deductions for repeat offenders are an alternative.
Is returns scoring legal in the EU?
It depends on the design. Personal data must be limited to what is necessary for a stated purpose, and the GDPR lets individuals refuse decisions based solely on automated processing with significant effects, with human intervention and a right to contest as safeguards. Seek legal advice for your markets and keep a human decision-maker involved.
One edition every weekday morning. Read in five minutes. Free for industry professionals.
SOURCES
- Appriss Retail: Retailers prepare for increased threat of wardrobing, counter fraud with AI
- Business Wire: Appriss Retail Benchmark Report Uncovers $796B in Total Retail Loss
- FashionNetwork UK: Retailers are curbing serial returners, report
- GDPR-Info: Article 5, principles relating to processing of personal data
- GDPR-Info: Article 22, automated individual decision-making, including profiling
