A Fair Process for Marketplace Disputes

A buyer and seller disagree about whether an item matched its listing and what remedy is fair. They trade messages, each sure the other is being unreasonable. The marketplace can leave them to work it out, decide the claim itself, or offer a process between those choices.

The marketplace may not be able to tell from those messages alone what was promised or whether the item matched the listing. Before asking whether AI can help, it is worth asking what a fair process would need: separate accounts, clear rules, a careful look at the available evidence, and a way to challenge the outcome. Some disputes may be handled with a straightforward return policy or human support. AI is only worth considering if it helps with a specific part of that process without taking responsibility away from the people who run it.

Start with the accounts

An exchange about an order can escalate quickly. Each new message can feel like building a case against the other. Before comparing evidence or suggesting a settlement, a marketplace facilitator could let each person describe the incident independently. That simple step gives both sides room to explain what they saw without first reacting to the other person's version.

If people find it useful, an AI system could reflect each account back in clearer terms, separating what a person says happened from what they infer about the other's intentions. The person could correct the reflection privately before deciding whether to share it. For example, a buyer might say an item did not match its description and ask for a refund. A reflection could restate that concern and let the buyer correct anything it missed. The seller could explain how they described the item and what they believe was agreed.

That reflection may help clarify what each person is asking for, but it can also distort or soften a report. Neither account should be treated as established because it is calm or well phrased. The process should not make the person who feels harmed responsible for being patient, or require anyone to agree, forgive, or meet. Sometimes the dispute is about a practical remedy, not a relationship that both sides want to repair.

Weigh the evidence and the rules

If the accounts still conflict, the marketplace needs a way to weigh them. It could ask both people for incident reports before either sees the other's statement, so one account is less likely to shape the other. This is not fully blind: a reviewer still has to compare the reports and any supporting material.

For a simple case, a clear policy and a human review may be enough. If the material is difficult to organize, AI might help lay out each claim beside the available evidence, such as the listing, transaction records, and messages. It could check which marketplace policies apply and offer possible next steps, such as a return, a partial refund, or referral to a human reviewer. Those are suggestions to consider, not an answer the system is entitled to issue.

Any such tool should distinguish evidence from interpretation, point out what is missing or disputed, and explain which policy informs each suggestion. It should not flatten anger into neutral language as if emotion made a report less credible. Photos can be incomplete, messages can lack context, and marketplace policies can be unfair. An orderly comparison is still a judgment about what counts; if a checklist or trained staff can do the job more clearly, there may be no reason to add AI.

Keep people in control

Each person should be able to correct the record, reject a suggestion, or ask for a human review independent of the first decision. The marketplace should explain its rules before a transaction and, if it uses AI, say what information the system considers. Private reports should not be shared or reused without permission.

Some cases call for a pause in messaging or help from a facilitator; others need a clear boundary, especially if someone feels unsafe or has been repeatedly targeted. Mediation should be voluntary. A person should not have to contact the other party to make a report, and account restrictions should be proportionate to the conduct rather than used as a substitute for resolving a transaction claim.

As The Architecture of Consensus argues, a fair process needs a path from disagreement to a decision and a record people can understand. Here, that path might end in a refund, another agreed remedy, or a human decision. AI might help organize a difficult case, or it might add little beyond another layer between people and a remedy. The first measure of a good process is not how much automation it uses, but whether both people can understand and challenge what happens.