New Chargeflow docs. Everything for merchants, platforms, and the API in one place.
ResourcesConcepts

Evidence generation

How Chargeflow builds a dispute response: written from scratch by multiple specialised agents rather than filled into templates, QA'd by an AI agent, aligned to card scheme rules, and tuned by the outcome of every win and loss.

Chargeflow writes each dispute response from scratch. There is no template with blanks in it.

That is the difference worth understanding before you evaluate any dispute vendor, because it decides what happens on the disputes that are not typical, and those are the ones you lose.

How it is built

StageWhat happens
Structure firstThe evidence structure is chosen for the reason code and the scheme before any prose exists.
Specialised agentsAgents trained per dispute type, per vertical, and per data-collection type each write the part of the claim they own.
CompositionA final agent composes and refines the parts into one coherent claim rather than a stack of sections.
BIN-driven optimisationThe issuer's BIN informs how the case is presented, because issuers do not all weigh evidence the same way.
QA by an AI agentFacts, card-scheme rules, and structure are checked before submission. See How our AI works.
Feedback-drivenWins and losses are A/B fed back, so the approach for each reason code keeps moving.

Multi-language support means the claim can be produced in the language the issuer works in.

Chargeflow versus template-based tools

ChargeflowTemplate-based providers
How the claim is writtenFrom scratch by multiple agentsPre-written template, AI fills the blanks
Adapts to the disputeYes, per reason code, vertical, and available dataLimited to what the template anticipated
Quality controlAI QA agent checks facts, rules, and structureUsually none beyond validation
Scheme alignmentA dedicated component aligns to card scheme regulationsStatic, updated when the vendor updates the template
Improves over timeEvery win and loss feeds backOnly when someone rewrites the template

The bottom line: each case is unique and the output mimics what a chargeback analyst would produce, rather than what a form letter produces.

What this means for you

Send the data, not the argument. Your job is context: the order, the conversation, the usage, the fulfilment record. Deciding which of it wins this particular dispute is the engine's job. Send everything you have; the enrichment engine evaluates what strengthens the case and leaves out what does not. See Enrich a dispute.

Unusual disputes are the point. Any tool handles a clean "item not received" with a tracking number. Value shows up on the dispute where the tracking number is missing but the customer's account history, IP, and support thread together make the case.

You can still review. Platforms can run co-pilot, where evidence is generated and a human approves before submission. See Platforms overview.

Next step

Background reading: Representment and evidence for the mechanics of a representment, and How recovery works for where generation sits in the pipeline.

Was this page helpful?

On this page