Agentic AI could strengthen financial-crime prevention, but governance, cost hurdles limit adoption: McKinsey

Agentic artificial intelligence (AI) could reduce the time banks spend on know-your-customer (KYC) processes by 30 to 70 per cent, improving efficiency in financial-crime prevention, but high human oversight costs and challenges related to governance and accuracy continue to hinder its wider adoption, according to a report by McKinsey & Company.

Agentic artificial intelligence (AI) could reduce the time banks spend on know-your-customer (KYC) processes by 30 to 70 per cent, improving efficiency in financial-crime prevention, but high human oversight costs and challenges related to governance and accuracy continue to hinder its wider adoption, according to a report by McKinsey & Company. The report noted that banks “often assign 10 to 15 percent of employees to know-your-customer (KYC) and anti-moneylaundering (AML) activities alone.” Now institutions are now turning to agentic AI.”

However, financial institutions often lack adequate metrics to assess AI reasoning and agent behaviour, despite such evaluations being relatively inexpensive as they do not require labelled data. Citing an example, it said one bank conducted ten evaluations across its customer onboarding process, assessing AI-generated outputs on accuracy, consistency across repeated runs, processing time and cost. The bank set minimum performance thresholds of 80 per cent accuracy and 70 per cent consistency. Across published evaluations, measured consistency ranged from 72 to 100 per cent.

“On the weakest, roughly one run in four returned a different answer to the same input,” it said. It further highlighted that human supervision accounts for a significant proportion of the operational costs associated with AI-driven banking services.

“When an agent performs a customer service task in banking, token costs account for only 20 to 25 percent of variable-run costs, while human oversight accounts for 70 to 75 percent,” it said, adding, “60 percent of an agentic task’s cost is tied to refining answers, rather than producing the first answer.” McKinsey noted that optimisation could reduce these costs by more than half through measures such as using cheaper AI models for simpler tasks, eliminating duplicate operations, reducing data transfers between agents and shifting from checking every output to sample-based quality reviews.

However, it cautioned that reducing the frequency of quality checks in regulated banking functions requires careful consideration, as such decisions involve compliance and risk management rather than merely technological efficiency. The report emphasised that ensuring reliable oversight becomes increasingly challenging as banks expand the use of autonomous AI systems. “The reason that having a human in the loop is no longer sufficient is supervisory,” it said.

Overall, “In financial-crime prevention, agentic AI has moved from concept to production faster than most institutions expected—and the value is real. But we are still at the beginning of the process,” it said. (ANI)