Agentic AI in financial-crime prevention: Lessons from winners
By McKinsey & Company ·
Drawing on a year of agentic AI deployments at European banks, McKinsey sets out five decisions that separate banks scaling AI in financial-crime prevention from those stuck at one use case. Banks assign 10 to 15 percent of staff to KYC and AML; agent pipelines can halve source-of-wealth checks that take up to 35 hours manually, and productivity across engagements is about twice the baseline. Quality control is the costly part: about half of a €2.50 transaction-monitoring case.
Key figures
| Figure | Value | Source |
|---|---|---|
| Share of bank employees in KYC and AML activities | 10 to 15 percent | p. 1 |
| Manual source-of-wealth verification vs multiagent pipeline | up to 35 hours; around half the time | p. 1 |
| Marginal cost per transaction-monitoring case, one large European bank | approximately €2.50 before optimization, roughly half of it quality control | p. 8 |
| Measured run-to-run consistency across published evaluations | 72 to 100 percent | p. 8 |
| Productivity across engagements vs baseline | about twice | p. 13 |
- Europe