---
title: "Customer due diligence and AML with AI"
description: "An agent ranks anti-money laundering alerts by priority, with a readable reason, and prepares the file; the analyst decides, those responsible report."
url: https://www.joinx.it/en/industries/use-cases/due-diligence-aml-fraud
---

# Anti-money laundering alerts in order of priority, with the file ready for the analyst

Every day the monitoring system produces alerts that the analyst opens one by one, working out afresh each time who the customer is and what they have done.

## One alert at a time, in order of arrival

In the anti-money laundering team, the analyst receives alerts from the transaction monitoring system and, for each one, opens the customer's position, looking for master data and risk profile in core banking, the customer due diligence documents in the archive, the period's transactions and the notes on earlier alerts. For suspected fraud the work is similar, although the signals come from other systems. The queue follows the order of arrival, so a serious case waits behind others that will be closed with no further action, and the same reconstruction is repeated at every periodic review.

## The agent sorts and prepares, the analyst judges

In the example, a model ranks the alerts by priority and writes the reason beside each one in readable words, for example a transaction far removed from the profile the customer declared. The agent then gathers transactions, documents and earlier cases into the file, and checks the registers that customer due diligence requires. The analyst reads the file and decides whether to close the alert or investigate further, while reporting a suspicious transaction stays with the people who are responsible for it. The agent does not block transactions and does not close accounts.

1.  The model ranks the alerts and writes the reason for each
2.  The agent puts the file together and checks the registers
3.  The analyst decides whether to close or investigate further
4.  The suspicious transaction report stays with those responsible

Illustrative example

## Data, modules and responsibilities in anti-money laundering controls

| Item | In this case |
| --- | --- |
| Data | Alerts from the monitoring system, customer master data and risk profiles, customer due diligence documents, transactions, the outcomes and reasons of alerts already closed |
| Modules | Data Foundation, Data Intelligence, Agents, Governance |
| Who decides | The analyst on closing and further investigation; on reporting suspicious transactions, whoever is responsible for it; the head of the anti-money laundering function on the rules and thresholds |
| What we measure | The time taken to put a file together, alerts closed with a written reason and files sent back for missing evidence, against the starting point |
| AI Act risk class | Anti-money laundering and fraud do not appear among the high-risk uses in Annex III that concern our industries, namely credit to individuals, life and health premiums, and staff. We classify the case with risk and compliance before development, and with Governance we record the model, rules and approvals of every run |
| Where to start | With a Mapping, because these are controls the supervisory authority reviews and it is worth choosing with the anti-money laundering function where to begin, for example with the periodic review of customer due diligence |

## Without the outcomes of past alerts the model cannot be tested

Testing a model that ranks alerts takes cases with a known outcome, that is, alerts already closed with the decision and the reason on record. If the outcome sits only in the analysts' free-text notes, or if two systems give different risk profiles for the same customer, the first job is on the data. The same applies when the criteria for closing an alert change from one analyst to the next, because they have to be written down with the head of the function before building.

The case comes from the Banking & payments page.

[Data and AI for banks and payments](https://www.joinx.it/en/industries/banking-payments) · [All use cases](https://www.joinx.it/en/industries/use-cases)

## Let's reopen together an alert that has already been closed

With an analyst we reconstruct a closed alert, to see how much time the file took and which systems the data came from.

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