Today's companies are drowning in communication data. Every day, millions of emails, instant messages, voice recordings, and video calls flow through their systems. For most companies, however, this is mostly an annoyance, like trying to track down a specific file from a certain conversation. Yet for financial firms, this wave of communication data potentially contains something that could bury them. That’s because financial firms operate under strict legal regulations requiring them to both record and monitor all communications relating to the business. But when examining something like Microsoft Teams—which hosts discussions ranging from staff lunches to strategy for customer investments—trying to sort, separate, and monitor all these conversations becomes increasingly difficult. Equally, most financial institutions remain stuck in the past, relying on lexicon-based surveillance systems that generate overwhelming volumes of false positives while missing critical risks entirely. However, the emergence of AI has presented a new path forward, promising to revolutionize how financial institutions approach communication surveillance. Yet Luke Church, Senior Product Marketing Manager at Smarsh, notes that "there's only about 20 large banks that are currently using AI in their surveillance investigations." This highlights the scope for transformation. But beyond broad impact, it raises the question: how can AI bring efficiency to compliance efforts by filtering noise, identifying more risks, and actually saving more than the cost of a new system?
The Limitations of Traditional Surveillance Systems
Traditional lexicon-based surveillance systems are proving inadequate in spotting communication compliance issues. Changing language use—from abbreviations to emojis, to UC features that allow threaded conversations or quote replies—has expanded the contexts in which conversations take place. Yet lexicon systems, operating on simple keyword matching principles, flag any message containing predetermined terms regardless of context, intent, or relevance. This leads to the system both flagging irrelevant messages and missing important ones. The result is not only an incomplete compliance picture, but also days of work lost as compliance teams sift through hundreds of irrelevant messages, each requiring careful analysis to determine whether it represents a genuine risk. It’s clear that the old lexicon-based system of compliance no longer meets the demands of modern communication. As Smarsh shows, the way forward is through AI systems. Church explains:
"We've seen in some cases that customers are detecting three to five more times actual risks within their organization once they make that jump from a lexicon-based search to our AI-enabled models."
Bringing AI into communication surveillance is the first step toward smarter, more efficient compliance. The next challenge is understanding how these solutions are applied in practice, and how they can transform an organization’s approach to risk.
Intelligent Agent: Empowering Surveillance with Domain-Adapted AI
Recognizing these fundamental limitations in lexicon systems, Smarsh has developed the Intelligent Agent, an AI-powered surveillance solution that addresses both noise reduction and enhanced risk detection. The system represents a paradigm shift from reactive keyword matching to proactive contextual analysis, leveraging advanced AI to understand not just what is being said, but the intent and risk level behind communications. The Intelligent Agent operates on two primary capabilities. The first is sophisticated filtering, which Church describes as being able to "autonomously remove a lot of noise that customers are seeing, like disclaimers, market research, automated messages, marketing spam, and scams." This isn’t simple keyword exclusion; the system uses contextual AI to understand the nature and purpose of communications. This greatly improves the accuracy of flagged messages. The second capability focuses on enhanced risk detection through "proprietary, domain-adapted LLMs which have multilingual support right out of the box." These models have been specifically trained on financial services communication patterns and can identify suspicious activities regardless of language. "Customers are able to detect a lot more risks once they deploy this across their data landscape, and they're able to then convert that over into their native language to better understand the risk across global borders," Church noted. This means risks that previously flew under the radar are now being captured, giving compliance teams greater confidence. Equally, all these strengths can work across multinational companies with multiple languages in use. Plus, Smarsh's system uses a no-code user interface that allows compliance teams to test, evaluate, and optimize before deployment without requiring extensive technical expertise. As Church explained, "You're actually able to test and validate how the model is going to work for you before you put it into production with your data." This testing capability addresses a primary concern financial institutions have about AI implementation: the fear of deploying systems they don’t fully understand or trust.
From Compliance Burden to ROI
The features AI compliance systems offer deliver quantifiable benefits that translate directly into measurable return on investment. Compliance teams spend less time reviewing irrelevant messages, detect risks more effectively, and gain actionable insights that were previously buried in noise. Church illustrates what this looks like in practice:
