Compliance teams at financial institutions are losing ground fast. According to new research from Saifr, in partnership with A-Team Group, senior compliance and technology leaders at 16 firms including investment banks, brokerages, asset managers and hedge funds have identified a surveillance crisis and why AI is no longer optional.
The core issue is volume. Millions of electronic communications are generated every day across email, instant messaging, collaboration tools and mobile channels. Legacy surveillance platforms, built for a simpler era, are struggling to cope.
The result is a compliance function drowning in noise, with false positive rates at some firms reaching as high as 99%. One tier-1 global bank reported 7,000 alerts generated from 5.5 million messages in a single day, the overwhelming majority of which were irrelevant. Analysts are spending up to 60% of their time reviewing non-issues, leaving less capacity to identify genuine misconduct.
The problem runs deeper than volume alone. Legacy systems rely on lexicon-based detection, flagging communications based on keyword matches with no understanding of context or intent. A conversation about baseball in which someone mentions “stealing bases” could trigger a compliance review. These systems, respondents told researchers, are simply “not sophisticated enough” to assess the narrative surrounding an alert or connect it to related communications and trade data. The consequence is excessive triage, wasted resource and a weakened ability to catch real risks.
Architectural rigidity compounds matters further. Many legacy platforms cannot integrate modern AI tools without significant redevelopment. One French tier-1 bank reported that its existing system was too inflexible to support advanced AI capabilities, forcing an internal shift to a modular architecture built specifically for that purpose. Manual processes such as ongoing lexicon tuning absorb compliance resource that firms can ill afford to spend this way.
Saifr’s findings point to AI and large language models (LLMs) as delivering measurable results across all three problem areas. On false positives, firms deploying LLM-powered surveillance are reporting reductions of 30% to 40% or more. One institution reduced its alert volume from 900,000 to just 16,000 following LLM implementation. AI-driven intelligent filtering is also enabling systems to recognise and suppress known benign patterns before alerts ever reach a human reviewer, with some firms piloting bulk closure of low-risk alerts at scale.
For more insights, read the full story here.
Copyright © 2026 FinTech Global
Copyright © 2018 RegTech Analyst
