Handling returns and disputes, in numbers

When a €280 vintage garment changes hands between two strangers across European borders, subjectivities multiply. One user's "distressed vintage patina" is another's "irrecoverable fabric damage." In standard second-hand commerce, platform operators absorb these collisions through blunt financial instruments: blanket buyer refunds that penalize sellers, or rigid no-return policies that scare off prospective buyers.

During the third quarter of 2024, WEVONE's Tutus universe processed 14,280 completed apparel transactions across six European territories. Exactly 600 of those orders—4.2% of total transaction volume—triggered an official dispute before final balance settlement.

The Taxonomy of Second-Hand Friction

Disputes in peer-to-peer fashion do not distribute evenly. An analysis of our Q3 dispute logs categorizes the underlying operational triggers into four distinct clusters:

• Item Not As Described (INAD): 58.3% (350 cases). Micro-stains, unmentioned alterations, pulled threads, or fabric degradation concealed in original listing photography.
• Transit Logistics & Damage: 23.5% (141 cases). Parcel loss, moisture penetration in transit, or extended carrier delays exceeding the contract window.
• Fit & Sizing Mismatches: 14.2% (85 cases). Discrepancies between marked tag size and actual flat garment measurements. (Note: Fit dissatisfaction with accurate measurements is non-refundable under Tutus rules).
• Buyer Remorse & Fraud Claims: 4.0% (24 cases). Unsubstantiated claims filed after delivery verification.

In traditional marketplace architecture, resolving 600 subjective disputes requires human support operators spending an average of 22 minutes per ticket, consuming operational margin and creating payout friction that degrades seller retention.

The 72-Hour Escrow Containment Protocol

The foundation of dispute mitigation in Tutus relies on a synchronous financial hold: the 72-Hour Escrow Ledger. When a package scan marks an item as delivered by the postal carrier, the transaction does not immediately settle into the seller's withdrawable balance. Instead, funds sit locked in escrow for precisely 72 hours.

If a buyer flags an anomaly within this window, the ledger automatically freezes the payout trigger. The buyer must submit three macro photos of the disputed area alongside a standardized physical scale reference within 24 hours of opening the claim.

This immediate hold prevents the "payout-and-run" pattern common on un-escrowed platforms, while enforcing a strict operational timeline that prevents buyers from holding seller capital hostage indefinitely. If no claim is filed within 72 hours, the transactional escrow auto-releases the funds to the seller's wallet, closing the order permanently.

Worked Example: The JPG Mesh Top in Lyon

To understand how computational arbitration functions in practice, consider order #TT-8902, executed in August 2024. A buyer in Berlin purchased a 1994 Jean Paul Gaultier printed mesh top from a seller in Lyon for €280.

Four hours after parcel collection, the buyer opened an INAD dispute, uploading a macro photograph showing a 1.8cm thread separation along the left side seam. The buyer demanded a 100% refund and return shipping coverage. The seller countered that the garment was intact when packaged.

Rather than routing the ticket to a manual moderation queue, Mia's Vision Matching pipeline initiated an automated visual comparison:

1. Vector Alignment: Mia aligned the buyer's dispute photo against the six original listing photographs using seam geometry anchor points.
2. Contrast & Zoom Reconstruction: The vision model zoomed into frame 3 of the original listing (a medium shot of the left torso) and applied edge-detection filtering.
3. Pre-existing Defect Identification: The analysis identified a localized structural thread tension anomaly—the precise initial stage of the seam split—visible in the original listing photo under 3x digital amplification.

Because the flaw was visible in the original photography but omitted from the seller's text description, Mia determined shared responsibility. The system calculated a proportional resolution: a €45 partial refund drawn from the escrow balance to cover local tailoring repair, with the remaining €235 released immediately to the seller. Both parties accepted the algorithmic offer within 11 minutes of claim initiation. Neither shipped the item back; neither absorbed a complete loss.

The Human Fallback and Unresolved Blind Spots

Computational mediation has clear technological limits. Mia can parse visual pixel variance, but she cannot smell damp mildew, assess fabric stiffness, or detect structural elasticity loss in synthetic waistbands.

Of the 600 Q3 disputes, 61% (366 cases) were successfully arbitrated by Mia within four hours via photo match or logistics carrier API verification. The remaining 39% (234 cases) were escalated to human moderation.

Scent-based claims ("garment smells like heavy tobacco") constitute 12% of all escalated tickets. Because odor cannot be digitized, Tutus currently enforces a strict evidence heuristic: unless the seller has a documented history of scent disputes (which degrades their platform Contribution Score), scent claims without visible residue are ruled in favor of the seller, with the buyer receiving a platform credit if their account holds high trust history.

Another ongoing structural vulnerability is item-swapping fraud, where a buyer returns a different, damaged, or counterfeit item back to the seller. In Q3, Tutus logged four confirmed cases of return-swapping. To combat this, we are currently testing an experimental optical signature protocol in beta, requiring sellers of items valued over €300 to record a continuous 10-second video encoding unique tag serials and fabric weave identifiers prior to sealing the parcel.

The Cost of Trust

Peer-to-peer marketplaces cannot eliminate human conflict; they can only measure it accurately and minimize the friction required to resolve it.

By shifting dispute resolution from subjective argument to deterministic escrow states and visual verification, Tutus brought the average time-to-resolution down from 4.5 days to 6.2 hours in Q3 2024. The platform's total loss rate from unrecoverable dispute payouts stands at 0.14% of gross merchandise volume—a tax we track, publish, and actively target for reduction as our computer vision models iterate.