Shaun Hurst, Principal Regulatory Advisor, Smarsh
The UK Financial Intelligence Unit received 866,616 suspicious activity reports in 2024–25, according to its most recent annual report, slightly fewer than the year before. Over the same period, the value of assets denied to criminals rose 59% to £382.6 million. One £103 million case accounts for part of that increase, but the figures still show law enforcement getting more from a similar number of reports.
Some of that volume is defensive filing, and the incentive behind it is easy to understand. The threshold for suspicion is low. A nominated officer who fails to report faces personal criminal exposure, while one who reports too much faces none. As long as that holds, firms will keep filing when in doubt, so the more realistic aim is to improve what each report contains.
An investigator reading a SAR wants to know who is involved and why the activity looks criminal. Transaction monitoring handles the first question reasonably well, from account and counterparty data. It struggles with the second. A series of transfers just under a threshold is unusual. A message telling a colleague to keep each transfer below that figure shows what someone intended.
That kind of evidence usually sits in communications. Banks already capture large volumes of written and voice communications, partly to meet recording rules for trading staff and partly for their own investigations. Where an employee is involved, the decisive material is often already in the archive, but the analyst drafting the report seldom sees it. Communications surveillance and financial crime investigations tend to run as separate functions on separate systems, and few firms have a routine way to pass evidence from one to the other. Some of the most relevant conversations never reach the archive at all, since they took place on WhatsApp or a personal phone.
Measurement makes this worse. Teams are typically judged on alerts cleared and backlogs reduced, which rewards speed. Applying AI against those targets will mostly produce more alerts and longer narratives that say similar things. The UKFIU plans to use machine learning in its SARs Digital Service to analyse reports, and narratives like these give it nothing specific to work with.
A more useful measure is how many reports lead to action, but firms can only track that with feedback from law enforcement. The UKFIU publishes reporter booklets with case studies of SARs that led to investigations. They're useful, but they don't tell an individual firm which of its own reports made a difference. Where firm-specific feedback does arrive, it tends to stay with the central intelligence team, and the analysts who wrote the original narratives hear little about which ones helped.
Banks can act on both problems without waiting for law enforcement. Passing feedback to the people who drafted the reports costs almost nothing. Giving investigators controlled access to relevant communications when they build a case takes more work: privacy and tipping-off risk both need managing, and someone has to decide which conversations are relevant. Firms that already run communications surveillance have most of the infrastructure in place.