JD Sports has quietly rolled out a behavioural artificial-intelligence system that scores the likelihood that an online return is fraudulent, with internal figures indicating the tool now flags roughly one in 12 refund requests for additional scrutiny. The system, deployed across the sportswear giant’s UK e-commerce operation, automatically routes high-risk cases for manual review or outright refusal — placing the FTSE 100 retailer at the sharp end of a fast-growing and ethically fraught corner of retail technology.
The move reflects a wider scramble among British retailers to stem the spiralling cost of returns abuse, a category that ranges from “wardrobing” — wearing an item before sending it back — to organised fraud involving empty parcels and counterfeit swaps. But it also raises uncomfortable questions about transparency, fairness and the prospect of loyal customers being silently penalised by an algorithm they cannot see or challenge.
How the system works
According to people familiar with the deployment, the detector ingests a broad range of behavioural and transactional signals rather than relying on a single trigger. These include the frequency and value of a customer’s past returns, the proportion of orders returned, mismatches between delivery and billing data, the speed at which items are sent back, and patterns associated with known fraud rings.
Each return is assigned a risk score. Low-scoring refunds are processed automatically, as customers would expect. Those above a threshold are funnelled into a manual review queue, where staff may request the item be inspected before a refund is issued — or, in the most serious cases, decline the refund and flag the account.
The reported flag rate of around one in 12 — roughly 8% of online returns — is striking. Industry analysts say it sits well above the share of returns most retailers would privately attribute to outright fraud, suggesting the model is casting a deliberately wide net that will inevitably sweep up legitimate shoppers.
“A flag is not a verdict, and that distinction matters enormously,” said Dr Priya Anand, a researcher in algorithmic decision-making at the Centre for Digital Commerce Studies. “If 8% are flagged but only a fraction are genuinely fraudulent, then the system is by design routing thousands of honest customers into friction, delay and the implicit suspicion of having done something wrong.”
The economics driving the crackdown
The commercial logic is hard to argue with. UK retailers absorbed billions of pounds in returns-related losses last year, and the rise of free, frictionless online returns has made abuse easier than ever. Fashion and footwear are particularly exposed, with return rates on some product lines exceeding 30%.
For a high-volume retailer like JD Sports, even a modest reduction in fraudulent refunds translates into meaningful margin. Behavioural AI promises to do at scale what no human loss-prevention team could: assess every single transaction in real time.
- Wardrobing: returning worn items, often after a single use.
- Empty-box fraud: claiming a parcel arrived empty or with the wrong contents.
- Serial returning: habitually ordering multiples with the intent to return most.
- Organised abuse: coordinated rings exploiting refund policies at scale.
“The pressure on retail margins is real, and loss prevention has become a boardroom issue rather than a back-office one,” said Marcus Feldwick, a retail technology analyst at Brightgate Advisory. “The danger is that the metric these systems optimise for — money saved — is far easier to measure than the cost of alienating a good customer who quietly never shops with you again.”
Transparency and the risk to genuine shoppers
The core tension is that the customer rarely knows the scoring is happening. A shopper whose perfectly valid return is delayed or refused may have no idea an algorithm was involved, no clear explanation of why, and no obvious route to appeal.
Consumer-rights advocates argue this sits awkwardly alongside statutory protections. Under the Consumer Contracts Regulations, online shoppers have a 14-day right to cancel most purchases — a legal entitlement that an opaque fraud score should not override.
“The law gives consumers clear rights to return goods bought online, and a risk model cannot lawfully erase those rights,” said Eleanor Voss, a consumer-law specialist at Harewood Chambers. “Retailers are entitled to investigate suspected fraud, but they must be able to justify a refusal. A black-box score is not, on its own, a justification.”
There are also data-protection implications. Automated decisions that significantly affect individuals can trigger obligations under UK GDPR, including the right to meaningful information about the logic involved and, in some cases, the right to human intervention — precisely the kind of safeguards that behavioural scoring systems are designed to streamline.
An industry-wide trajectory
JD Sports is far from alone. A clutch of fast-fashion and electronics retailers have begun trialling similar tools, and a growing vendor ecosystem now markets returns-fraud scoring as a standard component of the modern e-commerce stack. The direction of travel is clear: returns are becoming a moment of assessment rather than a no-questions-asked entitlement.
What remains unresolved is governance. Few retailers publish their flag rates, fewer still their false-positive rates, and almost none offer customers a straightforward explanation when a refund is held back.
What this means
JD Sports’ deployment is a clear signal that behavioural AI has moved from fraud’s frontline into routine retail operations — and that the everyday act of returning a pair of trainers is now subject to algorithmic judgement. For retailers, the payoff is tangible: tighter margins and a credible defence against organised abuse. But the one-in-12 flag rate underlines the central risk. Without published accuracy figures, clear appeal routes and respect for existing consumer rights, these systems threaten to convert good customers into collateral damage, all behind a screen of commercial confidentiality. As more retailers follow JD’s lead, the pressure will grow for regulators and the industry to define where loss prevention ends and unfair, opaque profiling begins.
Photo by İsmail Güngör Gedik on Pexels
