Fraud-Fighting Software Handles Routine Alerts, but People Make the Big Calls

A software company that helps banks and payment firms fight financial crime has changed how it uses AI. Flagright co-founder Madhu Nadig says AI agents should handle routine, lower-risk alerts. People should make the decisions that change a customer's financial life, such as closing an account. He wrote about the shift in a PYMNTS eBook, "Payments Leaders Rethink What Works in 2026."

What changed

At the start of the year, Nadig believed a person had to approve every alert the AI touched. The AI gathered evidence, summarized the case and drafted the write-up. An analyst then clicked to close or escalate it. That made it easy to show who was accountable. But on the thousandth low-risk alert in a queue, he says, a click carries very little judgment.

The early results

Nadig points to two customers. Both figures come from his own account, and the companies are not named.

• A payments company cut the time spent writing case narratives by 60% after AI took over documentation.
• A wealth management firm with more than 1 million customers cut the time to file suspicious activity reports (SARs) by 25%.

How oversight is checked now

Instead of a click on every alert, Flagright now watches override rates. That is how often investigators overrule the AI. If the number rises more than expected, the AI's performance is slipping. Because the AI logs every step, quality reviewers can also look at every case. Most programs today check a sample of 5% to 10%.

Nadig compares each AI agent to a new hire. It follows the institution's own procedures, logs what it does, and a person controls the outcome of every investigation.

Where people stay in charge

People keep the decisions with the highest stakes: account closures and SAR filings. Nadig also says regulators have stopped resisting AI in this work. FinCEN and the European Union's new anti-money laundering authority now encourage advanced technology. Accountability for every AI decision still rests with the institution.

What to ask

If your bank or business uses AI to check for fraud, two questions are worth asking. Which decisions does a real person still make? And how often is the AI itself checked? The material does not say how widely this approach is used beyond Flagright's customers.