Why the next breakthrough in financial crime investigations won’t be about replacing investigators, but preserving how the best ones think.

By Daniel R. Sigmond
Founder, Special Agent Advisory Group LLC | Retired FBI Supervisory Special Agent

I spent more than 20 years with the FBI, much of that time investigating financial crimes, securities fraud, and complex schemes involving individuals who went to extraordinary lengths to conceal what they were doing.

One thing I learned early in my career is that having more information doesn’t necessarily mean you understand what happened.

You can have thousands of financial records, emails, transaction histories, and investigative reports sitting in front of you and still miss the most important piece of the case.

Sometimes the breakthrough comes from a single transaction that doesn’t make sense. Other times, it’s a relationship between two people that nobody thought to examine. And occasionally, it’s recognizing that the theory you’ve been pursuing is wrong.

That’s investigative judgment. And after decades of watching technology evolve, I believe we’re approaching a point where artificial intelligence could fundamentally change how that judgment is developed, preserved, and applied.

But not necessarily in the way most people are discussing it.

We’re Asking AI to Do the Easy Part

Today, much of the conversation around AI in financial crime focuses on efficiency.

Can it review transactions faster? Can it reduce false positives? Can it automate suspicious activity reporting? Can it identify patterns that traditional monitoring systems miss?

Those are legitimate questions, particularly for financial institutions dealing with enormous transaction volumes and increasingly sophisticated criminal activity.

And AI is already proving useful.

Machine learning can identify unusual transactional behavior. Graph analytics can uncover relationships between accounts, individuals, and corporate entities. Generative AI can help investigators review documents, organize evidence, and summarize complex information.

But here’s my concern.

We’re spending a tremendous amount of effort teaching technology how to process information, while paying considerably less attention to teaching it how experienced investigators evaluate that information.

Those are two very different things.

Identifying an unusual transaction is relatively straightforward compared with determining whether it represents criminal activity, understanding its significance, and deciding what to investigate next.

An alert is not a conclusion. A correlation is not proof. And an AI-generated narrative, no matter how convincing it sounds, isn’t evidence.

Experienced investigators know the difference.

What Makes a Good Investigator?

Throughout my FBI career, I worked with some exceptionally talented investigators. They didn’t all approach cases the same way, but the best shared certain characteristics.

They were curious. They questioned assumptions. They understood that people lie, records can be misleading, and the most obvious explanation isn’t always the correct one.

They also knew when to change direction.

I’ve seen investigations where the initial evidence appeared to point toward one explanation, only for additional information to reveal something entirely different.

That’s why good investigators don’t simply look for evidence that supports their theories. They actively look for information that might disprove them.

Think about what that means for financial crime detection.

A transaction monitoring system might flag a series of payments involving multiple companies and foreign accounts. An experienced investigator wants to know much more.

Who controls those companies? What is the legitimate business purpose of the payments? Do the transactions make economic sense? Are there undisclosed relationships? What information would support an innocent explanation? What facts would contradict it?

And perhaps most importantly: What aren’t we seeing?

These questions reflect experience, not simply access to data.

That experience is accumulated over years of investigations, mistakes, successes, interviews, financial analysis, and decisions made with incomplete information.

Unfortunately, much of it walks out the door when an experienced investigator retires.

I know something about that transition myself.

The Knowledge We’re Losing

When I retired from the FBI in January 2026, I began thinking more seriously about something that had concerned me for years.

What happens to the collective knowledge of investigators after they leave?

The FBI, financial institutions, intelligence agencies, and other investigative organizations invest enormous resources developing experienced professionals.

Over time, those individuals develop instincts and decision-making frameworks that are difficult to capture in training manuals or standard operating procedures.

We document investigative results. We preserve evidence. We write reports. But we rarely capture the full reasoning process behind the decisions that produced those results.

Why did an investigator pursue one lead instead of another? What made a particular transaction stand out? What caused someone to question a seemingly credible explanation?

And what lessons came from the investigations that didn’t go as expected?

That knowledge has tremendous value.

I believe AI presents an opportunity to preserve at least some of it, provided we’re willing to approach the technology differently.

Teaching AI How Investigators Think

Imagine a financial institution with an AI-assisted investigative system that does more than generate alerts.

When suspicious activity is identified, the system helps an analyst evaluate competing explanations. It identifies missing information, suggests relevant lines of inquiry, and explains why particular relationships or transactions deserve additional scrutiny.

Rather than simply telling the analyst that something looks suspicious, it helps the analyst understand what needs to be established before reaching a defensible conclusion.

Now imagine that the system’s investigative framework has been developed with input from experienced FBI agents, forensic accountants, prosecutors, compliance professionals, and financial-crime investigators.

Not just their reports or final conclusions, but the reasoning behind their decisions.

That’s where I see an opportunity.

Of course, capturing investigative judgment is much more complicated than loading case files into a large language model and expecting it to think like a seasoned investigator.

It would require structured interviews with experts, carefully developed investigative scenarios, rigorous testing, and continuous evaluation.

It would also require acknowledging that experts sometimes disagree, make mistakes, or rely on assumptions that don’t hold up.

The goal shouldn’t be to reproduce those mistakes at scale.

It should be to identify the reasoning principles that make investigations more effective and translate them into tools that help people make better decisions.

That’s a very different ambition from simply automating another compliance workflow.

The Investigator Must Remain Accountable

I don’t believe AI should independently decide that someone has committed a financial crime.

Nor do I believe an investigator should accept an AI-generated conclusion without understanding how it was reached.

In my experience, investigative credibility depends on being able to explain what you know, how you know it, and what evidence supports your conclusions.

That standard doesn’t disappear because technology is involved.

AI systems can produce incorrect information. They can misunderstand relationships, overlook context, and present speculation with an alarming degree of confidence.

In financial crime investigations, those errors have real consequences.

An incorrect conclusion can damage a person’s reputation, misdirect an investigation, create regulatory exposure, or cause an institution to overlook actual criminal activity.

That means any serious application of AI must include human oversight, reliable data, documented reasoning, and the ability to challenge the system’s findings.

If an investigator cannot distinguish between verified evidence and an AI-generated inference, the technology has created a new risk rather than solved an existing one.

Financial Crime Doesn’t Respect Organizational Boundaries

There’s another aspect of this discussion that deserves attention: public-private partnerships.

During my FBI career, including assignments involving national cyber initiatives and engagement with the private sector, I saw firsthand how important these relationships are.

Criminal organizations don’t organize their activities around the reporting structures of financial institutions or the jurisdictional boundaries of government agencies.

They exploit the gaps between them.

A bank may see one piece of suspicious activity. A cryptocurrency platform may see another. A technology company may possess information that helps explain the connection. Law enforcement may have intelligence that changes the significance of all three.

No single organization necessarily sees the complete picture.

AI could help connect those pieces more effectively, but technology alone won’t solve the problem.

We still need trusted relationships, appropriate legal frameworks, responsible information sharing, and people who understand how to work across organizational boundaries.

I’ve always believed that effective public-private partnerships are built on more than exchanging information. They’re built on understanding what each organization can contribute and then turning that understanding into action.

AI should strengthen that process, not become another isolated system operating inside an institutional silo.

Where I Believe We’re Headed

Since leaving the FBI and establishing Special Agent Advisory Group, I’ve become increasingly interested in how we can use emerging technology to preserve expert investigative judgment.

It’s a concept I’m actively exploring, and I believe it has applications well beyond financial crime.

But financial crime is a logical place to start.

The volume of information is enormous. The methods criminals use continue to evolve. And the consequences of getting an investigation wrong can be substantial.

Financial institutions will understandably continue investing in AI systems that improve efficiency and reduce costs.

They should.

But I would encourage their leaders to ask a more important question:

Are we simply making our existing processes faster, or are we actually making our investigators better?

Those are not necessarily the same thing.

A system that processes twice as many alerts isn’t much of an advancement if it doesn’t help identify the activity that truly matters.

Likewise, a system that produces polished investigative summaries isn’t particularly useful if the underlying reasoning is flawed.

The real opportunity is to combine the analytical capabilities of AI with the judgment, experience, and skepticism of seasoned investigators.

That requires treating investigative expertise as something worth preserving, studying, and systematically incorporating into future technology.

After more than two decades with the FBI, I can tell you that the best investigators aren’t necessarily the people with the most information.

They’re the people who know what to do with it.

If we can use AI to preserve and extend that capability, we won’t simply change how financial crime is detected. We’ll change how the next generation of investigators learns to think.

About the Author

Daniel R. Sigmond is a retired FBI Supervisory Special Agent and founder of Special Agent Advisory Group LLC. His career includes more than 20 years with the FBI investigating complex financial crimes, securities fraud, and cyber-enabled offenses, along with leadership assignments involving national programs and public-private partnerships. He now advises organizations on financial crime, investigations, strategic risk, and emerging technologies.