Blog

Blending human expertise with AI for defensible surveillance decisions

Written by Kirke Cushing | Aug 13, 2025 1:30:00 PM

In today's financial world, compliance teams are drowning. There's no other way to put it. Between the myriad chatting/texting apps and traditional emails, the volume of daily communications requiring review has exploded beyond what any human team can realistically handle. Yet regulatory expectations haven't decreased—they've intensified.

Here's where things get interesting. While artificial intelligence (AI) offers a lifeline for processing these massive data volumes, we're seeing some firms make a critical mistake: they're either going all-in on AI automation or completely avoiding it out of fear. Both approaches miss the mark.

The sweet spot? A human-in-the-loop approach where AI handles what it does best—data aggregation and pattern detection—while compliance officers and supervisors provide the nuanced judgment that only humans can deliver. This isn't just about efficiency; it's about creating defensible surveillance decisions that regulatory bodies will respect.

Why pure automation falls short

Let's be honest about AI's limitations. Sure, it can flag thousands of messages containing suspicious keywords in seconds. But can it tell the difference between an advisor discussing a client's "confidential estate planning" and someone sharing material non-public information? Not reliably.

FINRA Rule 3110 makes this crystal clear—firms must maintain supervisory systems that are "reasonably designed" to achieve compliance. Notice the emphasis on "reasonably designed." This means understanding how your tools work and ensuring their outputs align with your legal obligations. The SEC's recent Roundtable on AI emphasized that while AI can enhance surveillance capabilities, human validation remains essential to interpret findings and avoid false positives.

Consider this real-world scenario: An AI system flags frequent mentions of "gift cards" in an advisor's communications. Without human context, this might trigger a gifts and entertainment policy violation alert. But a human reviewer quickly recognizes these are legitimate discussions about client holiday bonuses structured as gift cards—completely compliant and properly documented. That's the difference between smart automation and intelligent oversight.

Building blocks of effective human-AI collaboration

Here's how some leading firms are structuring their surveillance programs to leverage both AI efficiency and human expertise:

1. AI as the first line of defense

Some smart firms use AI to handle the heavy lifting: scanning communications across all channels, identifying patterns that might indicate violations, and prioritizing alerts based on risk levels. Modern AI tools can process natural language in real-time, detecting not just keywords but sentiment, context, and relationship patterns.

For AML and KYC compliance specifically, AI can excel at:

  • Flagging unusual communication patterns that might indicate suspicious customer activity
  • Identifying conversations about account opening procedures that deviate from standard KYC protocols
  • Detecting discussions of cash transactions that approach reporting thresholds
  • Spotting references to high-risk jurisdictions or sanctioned entities

2. Humans as the final arbiters

But here's where human expertise becomes irreplaceable. Compliance officers and supervisors need to:

  • Investigate flagged communications within proper business context
  • Assess intent behind potentially problematic language
  • Apply firm-specific policies and procedures to ambiguous situations
  • Make escalation decisions based on regulatory requirements

Think about customer due diligence communications. An AI might flag a conversation where an advisor asks detailed questions about a client's source of funds. A human reviewer can quickly determine whether this represents proper enhanced due diligence for a high-risk client or potentially inappropriate prying that could damage the client relationship.

Expanding beyond traditional surveillance: AML/KYC integration

One area where human-AI collaboration proves especially valuable is integrating AML and KYC compliance into communications surveillance. Traditional approaches often treat these as separate functions, but savvy compliance teams are recognizing the connections.

Enhanced customer due diligence monitoring

Modern surveillance systems should consider monitoring communications for:

  • Discussions about customer identity verification processes
  • Conversations regarding beneficial ownership requirements
  • References to politically exposed persons (PEPs) or high-risk customers
  • Communications about source of funds or wealth verification

Here's where human judgment becomes crucial. An AI system might flag every mention of "cash deposit" as suspicious. But experienced compliance professionals can distinguish between legitimate discussions about customer banking habits and potentially problematic patterns that warrant further investigation.

Sanctions and watchlist screening

AI tools excel at scanning communications for names, entities, and jurisdictions appearing on sanctions lists. However, humans are essential for handling:

  • Name variations and transliterations
  • Contextual references that might not trigger exact matches
  • Determining whether references represent actual business relationships or general discussions
  • Escalating potential matches through proper channels

Marketing communications: Where precision matters most

FINRA Rule 2210 creates specific obligations for reviewing marketing communications, and this area particularly benefits from human-AI collaboration. The stakes are high—marketing violations can result in significant regulatory action and reputational damage.

AI's role in marketing review

Artificial intelligence can dramatically speed up initial marketing review by:

  • Scanning for prohibited language and exaggerated claims
  • Checking that required disclosures are present and properly placed
  • Identifying performance claims that need substantiation
  • Flagging testimonials that might violate current regulations

Human expertise in context and compliance

However, marketing compliance requires nuanced judgment that only humans can provide:

  • Determining whether comparative claims are fair and balanced
  • Assessing whether disclosures are sufficiently prominent
  • Evaluating whether communications could mislead the average retail investor
  • Ensuring marketing materials align with firm's current registration status and approved business activities

For example, an AI tool might flag every mention of "tax benefits" in marketing materials. A human reviewer can quickly determine whether these references include appropriate disclosures about tax advice limitations and individual circumstances—critical distinctions that could mean the difference between compliant marketing and regulatory violations.

White paper → Considering AI solutions for your business? Ask the right questions.

Addressing ethical concerns around AI reliance

The compliance community rightly worries about over-reliance on AI systems. AI systems trained on biased or incomplete data may produce skewed results, potentially creating regulatory blind spots or unfairly impacting certain client populations.

Maintaining transparency and accountability

Leading firms address these concerns through:

  • Clear Documentation: Maintaining detailed records of how AI models operate, including data sources and decision logic
  • Regular Testing: Continuously monitoring AI outputs for accuracy and fairness, updating models as language and business practices evolve
  • Human Oversight: Ensuring every AI-generated alert receives human review before any action is taken
  • Escalation Procedures: Creating clear pathways for complex cases that require additional expertise or management involvement

Preserving fiduciary obligations

Remember, investment advisers and brokers must uphold fiduciary obligations regardless of what technology they use. The SEC's recent guidance makes clear that AI tools cannot override these duties, even if they streamline workflows. This means human judgment remains essential for:

  • Assessing whether flagged communications represent actual violations
  • Determining appropriate remedial actions
  • Ensuring that surveillance activities don't interfere with legitimate client service
  • Balancing compliance obligations with client confidentiality requirements

Practical implementation: Seven steps to help achieve success

Ready to build a defensible human-AI surveillance framework? Here's a practical roadmap:

  1. Start with Clear Governance: Assign specific ownership for AI tool implementation and output validation. Someone needs to own this process completely.
  2. Invest in Training: Your compliance team needs to understand both the technology and its limitations. Regular training ensures staff can effectively interpret AI outputs and recognize when human judgment is required.
  3. Build Robust Audit Trails: Document everything—how alerts were generated, who reviewed them, what actions were taken, and why. Regulators will want to see this paper trail.
  4. Establish Escalation Procedures: Create clear protocols for handling ambiguous cases, potential violations, and system malfunctions. Know who makes final decisions and how they're documented.
  5. Regular System Testing: Don't set it and forget it. AI models need regular updates to stay current with evolving language patterns and regulatory requirements.
  6. Maintain Human Review Standards: AI should enhance, not replace, your existing review procedures. Every AI-generated alert should receive meaningful human analysis.
  7. Stay Current with Regulations: The regulatory landscape for AI in financial services is evolving rapidly. Subscribe to FINRA notices, attend industry conferences, and maintain relationships with your legal counsel.

The competitive advantage of getting this right

Firms that successfully implement human-AI collaboration in surveillance gain several significant advantages:

Scalability Without Sacrifice: Handle growing communication volumes without proportionally increasing compliance costs or compromising review quality.

Regulatory Confidence: Demonstrate to regulators that your surveillance program combines technological efficiency with human judgment—in line with what they want to see.

Risk Reduction: Catch more potential violations while reducing false positives that waste precious compliance resources.

Operational Excellence: Free up compliance professionals to focus on high-value activities like trend analysis, policy development, and staff training rather than manual review of routine communications.

As one compliance director at a major regional firm told me recently: "AI doesn't replace our judgment—it amplifies it. We're finding risks we would have missed and eliminating noise that used to consume our entire day."

Looking ahead: The future of collaborative intelligence

The financial services industry is still in the early stages of understanding how to effectively blend AI capabilities with human expertise. But the direction is clear: firms that master this collaboration will have significant advantages in managing compliance costs, reducing regulatory risk, and maintaining operational efficiency.

The key is remembering that AI is a tool, not a replacement for human judgment. FINRA's StratIntel team puts it well—the goal is to "uncover risks and opportunities" through collaborative intelligence, not to remove humans from the equation.

By maintaining a human-in-the-loop approach while leveraging AI's processing power, firms can build surveillance programs that are both defensible to regulators and practical for daily operations. That's a combination that helps protect both your firm and your clients—which is exactly what effective compliance should accomplish.

Remember, the firms succeeding in this space aren't just buying AI tools—they're thoughtfully integrating them into human-centered compliance programs. That integration, done right, creates surveillance capabilities that neither humans nor AI could achieve alone.

If you're thinking about integrating AI into your firm's risk and compliance programs, I encourage you to read this white paper: Considering AI solutions for your business? Ask the right questions. 

 

 

The opinions provided are those of the author and not necessarily those of Saifr or its affiliates. Saifr and any other third parties are independent entities and not affiliated. Mentioning them does not suggest a recommendation or endorsement by Saifr.

1209133.1.0