One of the most engaging discussions at our ENGAGE NYC event this year centered on a question many compliance and surveillance leaders are wrestling with right now: what does AI actually change inside surveillance operations?
Not in theory. Not in vendor presentations. But in the day-to-day reality of alerts, investigations, analyst workloads, governance and risk management.
This was the focus of a session on “Enhancing the Surveillance Team's Productivity with AI” which brought together Konstantinos Rizakos (General Manager, Compliance, NICE Actimize), Stan Yakoff (Founder, RegLabs.ai / Law Professor, Fordham Law School), Dave Frederick (Head of Americas Investment Compliance, Schroders) and Kerry Gendron (Head of Product, Surveillance Innovation, JPMorgan Chase).
What made the conversation particularly compelling was that it never drifted into hype. Instead, it focused on the practical tension firms are facing right now: balancing the enormous potential of AI with the operational, regulatory and human realities that come with implementing it.
AI Is No Longer Experimental. But It’s Also Not Magic.
One of the strongest themes throughout the discussion was the idea that AI is not some entirely new phenomenon that suddenly appeared overnight. Many of the concepts being discussed today have existed for years under different names: statistics, analytics, machine learning and data science.
What has changed is the speed at which organizations are now trying to operationalize these technologies across their businesses.
That distinction mattered because it immediately shifted the conversation away from “AI for AI’s sake” and toward something more useful: solving real operational problems.
Surveillance, after all, is not one giant workflow. It is a series of smaller, interconnected tasks: identifying risk, reviewing alerts, gathering information, investigating activity, documenting findings, escalating issues and maintaining consistency throughout the process.
And not every part of that workflow requires the same type of technology.
Kerry Gendron emphasized that firms are looking at how AI can identify risk more effectively and efficiently, while recognizing that “it’s not a one-size-fits-all.”
That mindset framed much of the conversation that followed.
The Biggest Productivity Gains May Come From Simpler Problems
One of the more surprising insights from the discussion was that some of the highest-return productivity gains firms are seeing are not necessarily coming from highly sophisticated AI deployments. In fact, several examples focused on much simpler operational improvements.
In fact, Stan Yakoff shared that some of the most effective solutions he had ever implemented had “nothing to do with AI.” Integrating portfolio manager notes directly into surveillance reviews, automating access to relevant news and standardizing alert documentation have all created massive efficiency gains because they reduced one of the biggest pain points analysts face every day: gathering information.
That observation resonated strongly because it reflects a reality many surveillance teams know well. Analysts often spend enormous amounts of time navigating systems, pulling context from different databases and piecing together fragmented information before they can even begin evaluating whether activity is genuinely suspicious.
That’s where AI is already creating meaningful value.
Rather than replacing analysts, firms are increasingly using AI to reduce the operational drag surrounding investigations. Dave Frederick discussed how AI can help consolidate information from multiple systems into a single view, making it easier for analysts to focus on judgment and decision-making instead of manual searching.
“Rather than the analyst having to check four or three different systems, [AI] can kind of look across all the systems and pull all the relevant data you need, have it at one place,” Frederick explained.
That distinction became important throughout the discussion. The real opportunity is not necessarily removing humans from the process. It’s helping them spend more time on the parts of the job that actually require expertise.
The Human Element Isn’t Going Away
As the discussion evolved, so did the central tension underneath it all: if AI becomes increasingly good at reviewing alerts and recommending outcomes, how much should humans rely on those recommendations?
That’s where the conversation became especially nuanced.
Examples were shared where AI — including targeted use of large language models — reduced false positives in communications surveillance. This led to a broader discussion of automation bias, or the tendency for people to over-trust machine-generated outputs. Participants agreed that as AI systems become more transparent in explaining why content is flagged as problematic, managers take on a new responsibility to monitor how often analysts disagree with the AI’s conclusions.
The concern is not theoretical. Regulators are already beginning to ask firms how often analysts override AI recommendations and whether humans are genuinely reviewing alerts independently or simply validating what the system has already suggested.
The consensus across the discussion was clear: human oversight remains essential.
“We always feel the AI tools are not there to replace compliance individuals,” Frederick said, “but rather enhance what they do.”
That point surfaced repeatedly in different forms. AI may help surface better signals, aggregate context faster and reduce manual effort, but accountability still sits with people. In highly regulated environments, firms cannot afford blind trust in automated systems.
Explainability and Governance Are Becoming Critical
As the conversation shifted toward risk and implementation, governance quickly emerged as one of the most important themes.
There was broad agreement that firms cannot rush into large-scale AI deployment without first building strong oversight frameworks around how models operate, how outputs are monitored and how decisions can ultimately be explained.
Yakoff warned about the dangers of overconfidence in black-box systems, particularly when organizations may not fully understand how certain outputs are being generated. The panel also discussed the reputational risks of flawed AI outputs and the importance of avoiding the kind of mistakes that end up making headlines.
That concern around explainability is becoming increasingly important as firms move from experimentation into production environments.
Legal teams, compliance leaders and business stakeholders all need visibility into how systems are functioning, how outcomes are validated and how organizations would defend those decisions under regulatory scrutiny.
One of the frameworks that resonated most strongly throughout the conversation was the idea of approaching AI adoption in phases: crawl, walk, run.
Start with smaller use cases. Build trust in the outputs. Test continuously. Put controls and monitoring in place. Then scale carefully over time.
That practical approach helped ground the discussion in operational reality rather than abstract ambition.
At several points, the panelists pushed back on the idea that firms should simply deploy AI everywhere possible. In fact, one of the strongest observations from the discussion was that the firms likely to succeed will not necessarily be the ones using the most AI, but the ones applying it thoughtfully to the right problems.
“The winners... are not going to be the firms who have deployed AI everywhere,” Yakoff said. “It’s going to be the firms that focus AI on the right problems.”
Surveillance Teams May Be Entering a New Era
By the final portion of the discussion, the focus shifted from improving existing workflows to something much bigger: could surveillance itself eventually operate differently?
The conversation explored whether future AI systems might move beyond reviewing alerts after the fact and instead continuously monitor communications, trading behavior, contextual signals and behavioral patterns in real time strongly enough to help prevent misconduct before it happens.
It was one of the more thought-provoking moments because it fundamentally reframed the discussion. The question stopped being “How do we make surveillance faster?” and became “How might surveillance itself evolve?”
At the same time, there was realism around the fact that bad actors evolve alongside technology. Even as AI systems become more capable, human judgment, skepticism and expertise will remain central to the process.
What is changing is the nature of the work itself.
Tomorrow’s surveillance teams may spend far less time manually collecting information and far more time validating AI outputs, investigating higher-risk activity, refining models and overseeing increasingly intelligent systems.
And that may ultimately be the biggest shift of all.
Final Takeaway
What made this discussion so compelling was not just the technology itself, but the realism surrounding it.
Nobody presented AI as a silver bullet. Instead, the conversation reflected what many firms are experiencing right now: AI can absolutely improve surveillance productivity, consistency and efficiency, but only when implemented thoughtfully, governed carefully and aligned to real operational challenges.
The organizations seeing the greatest success are not necessarily the ones moving the fastest. They are the ones who deeply understand their workflows, identify the right friction points and apply AI strategically where it can create meaningful value without sacrificing accountability or oversight.
And if there was one thing the discussion made clear, it’s that this transformation is only beginning.