Gift card marketplaces carry two costs that pull against each other. Fraudulent orders produce losses and fees. Controls that hold up legitimate orders produce delays and human reviews. Raise, which the case study says has processed more than $10 billion for about 6 million users across 1,000 brands, was paying both.
Stolen Gift Cards Convert to Cash Before Reviewers Catch Up
Gift cards are particularly vulnerable to fraud because they can be purchased digitally, transferred and redeemed before a disputed payment is identified. Fraudsters can use stolen payment credentials to purchase gift cards and convert them into goods or resale proceeds. Doriel Abrahams, head of risk at Forter, told PYMNTS in 2023 that a gift card is the online world’s version of cash. At food and beverage retailers, Abrahams said, gift cards were 5% of traffic but 50% of fraud. Raise saw the same pattern. Scammers bought a card, spent it, then disputed the charge, the case study said.
Raise’s in-house controls could not keep up. Reviews had grown to 15% to 18% of transactions, each taking 15 minutes to 24 hours and together consuming more than 10,000 hours a year. Customers waited longer for approvals. IP masking, proxies and location spoofing passed unflagged, according to the case study.
“Chargebacks were becoming a significant drain on our revenue,” Raise Risk Supervisor Dartanion Wells said in the case study.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
To close those gaps, Sardine tuned the rules Raise already ran, then trained custom machine learning models for gift card schemes specific to Raise, the case study said.
Across its platform, Sardine evaluates device fingerprints, IP and location data, proxy use, typing cadence and mouse movement, and feeds those signals into both a rules engine and models. A rule fires when one value crosses a threshold. A model weighs combinations, such as a new device, a masked IP and an odd order size together, that no single rule catches. The case study does not list which signals Raise’s deployment uses.
The approach is not unique to Sardine. ModCloth, which named gift card fraud its biggest threat, told PYMNTS in 2021 that it scores every order with its own models and has data scientists check whether screening is too tight.
Manual Reviews Fall to Fewer Than 3 in 1,000 Orders
At 0.25%, fewer than 3 of every 1,000 Raise orders now go to a human reviewer. At the earlier 15% to 18% rate, 150 to 180 of every 1,000 did. Wells said training one system once proved easier than training many reviewers many times. The case study does not say how many reviewers were redeployed, what the labor saving was worth, or the value of chargebacks avoided.
Manual reviews, chargebacks and false declines are distinct measures. The case study documents the first two. It does not say whether fewer legitimate orders are declined or whether sales rose. That gap matters: 47% of merchants estimate up to 5% of legitimate orders are wrongly declined, and 85% say their biggest fraud challenge is cutting friction for good customers while stopping bad ones, PYMNTS Intelligence found in its “Payments Orchestration Tracker.”
Still, the economics extend beyond chargebacks. Investigations require employee time, while delayed approvals can interrupt purchases. Reducing the number of transactions entering review can lower that workload without loosening fraud screening, as Raise’s 45% drop in chargebacks shows.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
Add as Preferred Source
