Compliance teams at financial institutions are buckling under the weight of outdated surveillance systems, and a new survey by A-Team Group suggests the breaking point has arrived.
Researchers interviewed senior compliance operations and technology leaders at 16 firms, spanning investment banks, brokerages, asset managers and hedge funds, to understand what is driving the accelerating shift toward AI-driven electronic communications (e-comms) surveillance.
Commenting on the findings, Saifr stated they paint a stark picture of an industry stretched thin. The explosion of e-comms platforms, from email and instant messaging to collaboration tools and mobile channels, has made monitoring for market abuse, insider trading and misconduct significantly more complex. Remote and hybrid working models have compounded the strain, as has growing regulatory pressure from authorities around the globe.
At the heart of the problem is the chronic inefficiency of legacy surveillance architecture, which Saifr identified as failing across three critical dimensions.
The most glaring issue is the prevalence of false positives. Legacy systems are generating alert rates that, in some cases, reach as high as 99%, meaning nearly every flag raised by the system turns out to be irrelevant. One tier-1 global bank reported receiving 7,000 alerts from 5.5 million messages in a single day, with the vast majority proving to be false positives. Analysts at affected firms report spending up to 60% of their working time triaging non-issues, time that could otherwise be directed at genuine risks.
The second challenge is a fundamental lack of contextual intelligence. Legacy platforms rely heavily on lexicon-based detection, a method respondents described as outdated and ill-suited to the nuances of natural language. Systems trigger alerts on standalone keywords with no reference to surrounding conversation, meaning a casual discussion about baseball, in which someone mentions “stealing bases”, could generate a compliance review. Without the ability to assess intent or narrative context, the quality of surveillance remains poor regardless of alert volume.
The third limitation is architectural rigidity. Many legacy platforms are simply too inflexible to integrate emerging AI capabilities, forcing firms to build parallel systems or undertake costly full replacements. One French tier-1 bank told researchers its legacy platform could not support advanced AI, prompting a full internal pivot to a modular, AI-ready architecture. Manual processes such as ongoing lexicon tuning add further drag, consuming compliance resource that cannot be redeployed elsewhere.
Against this backdrop, AI and large language models (LLMs) are gaining traction as credible solutions, not merely theoretical ones, it said.
On false positive reduction, the results are notable. Firms adopting LLM-based surveillance are reporting reductions of between 30% and 40% or more in alert volumes. In one striking case, an institution cut its total alerts from 900,000 down to 16,000 following LLM tuning.
AI is also enabling richer contextual analysis. Intelligent filtering can identify known benign patterns and remove them from the alert queue before human review. Advanced natural language understanding allows systems to distinguish between genuinely suspect communications and phrases that are merely superficially problematic. Some firms are now piloting AI-driven bulk closure of alerts where consistent benign patterns are detected across similar communications at scale.
On the architecture front, firms are investing in modular, flexible systems designed to absorb future AI advancements without major redevelopment cycles. One respondent highlighted that their new infrastructure had been specifically designed to benefit from “regular advances in AI”.
For more insights, read the full story here.
Copyright © 2026 FinTech Global
