---
title: "Disputes, chargebacks and complaints with AI agents"
description: "The back office finds the disputed transaction reconstructed by an agent, with the evidence gathered in time; authorised staff approve refunds and chargebacks."
url: https://www.joinx.it/en/industries/use-cases/disputes-chargebacks-complaints
---

# Reconstructing a disputed transaction before the deadline expires

When a customer disputes a charge, or a merchant has to respond to a chargeback, someone tracks down the transaction, authorisations and correspondence across several systems while the card scheme's deadlines run.

## The same searches, with every dispute

Disputes arrive through different channels: a customer's written complaint, a dispute opened through the card scheme, a merchant's report. They often end up in separate queues. The back-office operator looks for the transaction in the authorisation logs, retrieves the receipt or proof of delivery and checks whether the same customer has complained before, while the deadline set by the scheme rules draws closer. Sometimes a dispute is lost because a document arrived late even though the bank was in the right, and the same search is repeated on almost identical cases.

## The file arrives complete, and refunds and chargebacks stay with an authorised operator

In the example, the agent reads the dispute, classifies it by reason and reconstructs the transaction from the bank's systems, gathering in the file the evidence the rules require for that reason. It keeps track of the deadlines and moves the cases about to expire to the top of the queue. It then prepares the reply for the merchant, the card scheme or the customer, but does not send it, because a refund or a chargeback affects someone's account and is approved by an authorised operator.

1.  The agent classifies the dispute and reconstructs the transaction
2.  Gathers the evidence and ranks the cases by deadline
3.  The operator checks the reply and sends it
4.  An authorised person approves the refund or chargeback

Illustrative example

## Data, modules and responsibilities in a dispute

| Item | In this case |
| --- | --- |
| Data | Transactions and authorisations, correspondence with customers and merchants, complaint history, card scheme rules |
| Modules | Agents, Workspace, Company Brain |
| Who decides | The back-office operator on the reply; a refund or a chargeback is approved by an authorised person |
| What we measure | The time needed to reconstruct a case, disputes lost because evidence was missing or late, and the share of cases that still go to a person, compared with the starting point |
| AI Act risk class | The case does not assess the creditworthiness of individuals, the high-risk use in Annex III that concerns banks. If an assistant replies directly to the customer, it must make clear that it is an AI system. Before development begins, the risk and compliance functions confirm the classification with us |
| Where to start | With Deployment, because disputes usually have an owner and steady volumes, and every closed case is a case with a known outcome on which to test the agent |

## With few cases, or evidence outside the systems, it is better to wait

With a few cases a month, handled by someone who knows every merchant, the time saved does not repay building the agent. The same is true when the evidence, such as receipts or correspondence, sits in personal mailboxes or in archives the systems cannot reach. In that case the useful work starts with bringing it to where the bank can read it, and the agent comes later.

This sheet is part of the [use cases](https://www.joinx.it/en/industries/use-cases); for the other processes of banks and financial intermediaries, go back to [Banking & payments](https://www.joinx.it/en/industries/banking-payments).

## Let's start with a dispute that keeps recurring

With the back office we choose a type of dispute that needs the same searches every week, and look together at where it drags.

[Contact us](https://www.joinx.it/en/contact?settore=banche-pagamenti)
