How Disputes Work in AI-Powered Escrow

Most escrow disputes come down to one question: what was actually promised, and what actually arrived. Escrozon's process is built around answering that from the documentary record rather than from whoever argues more forcefully.

Opening a Dispute

Either party can open one during an active deal. You provide a reason, and the deal freezes immediately — no funds move, no further actions run until it resolves. Both parties are notified, and Phoenix AI posts an opening mediation message.

Freezing first matters. It removes any incentive to rush the other side into confirming while an argument is underway.

The Mediation Stage

Phoenix AI opens by reviewing the case file — the listing as it stood, the full chat history, and the deal timeline — then asks the party who filed to explain what went wrong, and puts clarifying questions to both sides one at a time.

The value here is structural rather than clever. Disputes usually arrive as a paragraph of frustration. Being asked "what did the listing say, and what did you receive" separates the substance from the feeling, and that separation is most of the work.

Evidence

Both parties can upload screenshots, PDFs, receipts, chat logs from elsewhere, code samples, and test output.

Phoenix performs forensic checks on what is submitted: whether images show signs of digital editing, whether timestamps are consistent with the deal timeline, and whether claims cross-reference against platform records. Each item gets a confidence score.

What makes evidence strong, in practice:

Contemporaneous beats reconstructed. A screenshot taken at delivery is worth more than one taken a week later.

Full context beats a crop. A cropped screenshot showing a number is weak. The same screen with the URL bar, the account name, and the surrounding interface is not.

Specific beats voluminous. Two screenshots showing the gap between listing and delivery beat thirty showing everything.

The listing text is the anchor. Almost every dispute resolves against what the listing promised, so quote it directly.

The Case Summary

When investigation completes, Phoenix generates a structured report: a case timeline, each party's position, an inventory of evidence with quality assessments, findings, confidence scores, and a recommendation.

The recommendation is advisory. It is explicitly labelled as such, because the decision is not the AI's to make.

The Moderator Decision

A human moderator reviews the case summary, the raw evidence, the complete chat history, and the listing description, then decides:

Release to seller — the delivery matched what was promised.

Refund buyer — it did not, materially.

Partial refund — delivery was incomplete rather than absent, or both parties contributed to the problem. This is more common than people expect, and it is usually the fairest outcome when a seller delivered most of what was agreed.

Why AI and Humans Are Split This Way

AI is good at the parts that are mechanical and tedious: reading a long chat history, checking timestamps, spotting an edited image, assembling a timeline. It does this consistently and without getting tired at message four hundred.

Humans are better at the parts that require judgement: whether a difference is material, whether someone acted in good faith and simply misunderstood, and what a fair split looks like when neither side is wholly right. Those are contextual questions, and they carry consequences for real people.

Handing the analysis to the machine and the decision to a person plays to both.

Timeline

Freezing is immediate and mediation begins in real time. Evidence collection usually takes a day or two, depending on how quickly both sides respond. Analysis follows, then moderator review. Most disputes resolve within a few days, with the largest variable being how fast the parties supply evidence.

How to Come Out of This Well

Respond promptly — silence is read as disengagement. Upload evidence when you file rather than waiting to be asked. Stay factual, because exaggeration on one point undermines your credibility on the others. Keep everything on-platform, since moderators can only weigh what is in the record. And accept a partial outcome where it is warranted; contested deals that end in a split usually reflect a genuinely split responsibility.

Frequently Asked Questions

Does Phoenix decide the outcome? No. It analyses evidence and produces a recommendation explicitly marked advisory. A human moderator decides.

Can I appeal? Contact support with anything genuinely new. A decision will not be revisited on the same evidence restated.

What if the other party never responds? The process continues. Non-response is itself part of the record and generally does not favour the party who is silent.

Is my evidence visible to the other side? Both parties see the case and can respond to what is submitted. Disputes are adversarial but not secret.

What happens to my account after a dispute? Nothing, for a legitimate dispute. Repeated bad-faith filings or confirmed fraud are a different matter and are handled separately.