Everlane

Return fraud has become a material operational risk in e-commerce, with industry data indicating roughly 9% of all returns are fraudulent and tactics growing increasingly organized. For Everlane, a direct-to-consumer fashion retailer built on radical transparency, the exposure was compounded by its catalog: many products share similar silhouettes, color palettes, and construction details, making item-swap detection difficult during visual inspection. Organized fraud communities specifically targeted the brand, exploiting mail-in return channels where items aren't verified until they reach the warehouse. Common schemes included FTID manipulation, empty-box returns, and swaps of lower-value or visually similar items from other brands — each generating a refund issued against merchandise never actually received.

Everlane, a Happy Returns partner since 2017, deployed a three-layer fraud prevention system combining physical item verification, behavioral AI, and computer vision. The first layer — the Return Bar® network — stations trained staff at drop-off locations to scan and verify every item at intake using a no-box, no-label model, blocking FTID manipulation and common mail fraud schemes before they enter the reverse logistics stream. The second layer uses behavioral risk scoring via partner Wyllo, which applies AI to detect unusual return patterns — multiple high-value returns, irregular drop-off locations, or unusually fast return cycles — and holds flagged refunds for review; fewer than 1% of returns are escalated. The third layer, Return Vision™, is a camera-based computer vision tool that compares photographed returned items against catalog images and product descriptions, flagging discrepancies in fabric, knit pattern, and construction detail. Audits complete within 2–3 days at Happy Returns Return Hubs, preserving a fast refund experience for legitimate customers.

The layered system delivered measurable financial protection at every tier. Behavioral risk scoring stopped $30,000–$40,000 in fraudulent refunds per month by intercepting FTID fraud before refunds were issued. Return Vision™ confirmed 18% of all flagged returns as fraudulent, with each caught instance preventing an average of $240 in loss from mismatched-item refunds. At the foundation, in-person Return Bar® verification reduced overall fraud by at least 85% compared to traditional mail returns. Operationally, fewer than 1% of returns require escalation, keeping the experience frictionless for the roughly 85% of Everlane shoppers who already choose Return Bar® drop-off over mail.

• 85% fraud reduction vs. mail returns (Return Bar® in-person verification)
• $30K–$40K/month in FTID fraud stopped via behavioral risk scoring
• 18% of flagged returns confirmed fraudulent by Return Vision™
• ~$240 average loss prevented per fraud instance caught

• A layered fraud stack — physical verification, behavioral AI, and computer vision — significantly outperforms any single control, especially for brands with visually similar SKUs.
• In-person item verification at drop-off eliminates the majority of fraud through both deterrence and process control, before items enter the reverse logistics stream.
• Behavioral risk scoring can isolate high-risk outliers at low volume (under 1% of returns) without degrading the experience for legitimate customers.
• Computer vision auditing is most effective as a final-layer tool for edge cases where manual inspection cannot reliably detect subtle item or material mismatches at scale.