Marketplace dispute resolution is a volume problem with a fairness requirement. A platform handling thousands of transactions per day will generate hundreds of disputes. Manual review of every dispute is expensive and slow, and inconsistency in resolution decisions creates perceived unfairness that damages seller relationships and buyer trust. AI handles the evidence gathering and initial assessment, reducing resolution time and improving consistency.

Dispute Types

Marketplaces encounter a limited set of dispute categories that drive most of the volume:

  • Item not received (INR): buyer claims order was never delivered
  • Item not as described (INAD): buyer claims item differs materially from listing
  • Defective item: buyer claims item arrived damaged or non-functional
  • Unauthorized transaction: buyer did not initiate the purchase
  • Seller non-fulfillment: order was placed but not shipped

Each category has different evidence requirements and resolution options. The triage layer classifies the dispute type from the submitted complaint text and routes it to the appropriate evidence collection workflow.

Evidence Collection

The dispute assistant automatically gathers available evidence before presenting anything to a human reviewer:

Order data: purchase date, item details, stated shipping date, tracking number, delivery confirmation status, item price.

Communication history: messages between buyer and seller through the platform, timestamps, any resolution attempts already made.

Seller history: dispute rate, fulfillment rate, account age, prior dispute outcomes.

Buyer history: dispute rate, purchase frequency, prior dispute outcomes.

Shipping carrier data: for INR disputes, the carrier tracking record including all scan events, delivery confirmation type (signature vs. doorstep), and any exception events.

This evidence package is assembled automatically. The human reviewer receives a complete dossier, not a bare complaint.

Policy Application

Each dispute category has resolution policy rules. For INR disputes with confirmed carrier delivery to the correct address, policy typically favors the seller unless buyer history suggests a pattern. For INAD disputes with photographic evidence that clearly shows a material difference from the listing, policy typically favors the buyer.

The assistant applies these rules to the evidence package and generates a policy-driven recommendation: resolve in favor of buyer, resolve in favor of seller, or escalate for human judgment because the evidence is ambiguous or the case falls outside standard policy.

The recommendation includes the policy rule applied and the evidence supporting it. This is the output the human reviewer evaluates - they are checking whether the rule and evidence are correctly applied, not starting from scratch.

Escalation Paths

Cases that require human judgment include: high-value disputes where the financial stakes warrant extra care, cases where buyer and seller histories are both clean and evidence is genuinely ambiguous, cases involving potential fraud by either party, and cases involving items in regulated categories.

Escalated cases go to a specialist queue with the full evidence dossier and the system’s assessment of why the case is ambiguous.

Seller and Buyer Communication

Resolution notices are generated from templates configured by the marketplace, with case-specific facts inserted. Both parties receive an explanation of the outcome. Sellers who lose disputes receive information about what evidence would have supported a different outcome - not to encourage gaming but to enable legitimate sellers to understand what documentation they need.

Dispute rates and resolution patterns feed back into seller risk scoring, influencing future order visibility and payout timing.