I remember sitting in a boardroom a few years back, listening to the head of compliance for a major bank. He was showing us their anti-money laundering (AML) system. It was a mess of if-then statements, a rigid, brittle system that was flagging thousands of false positives a day. His team was drowning in alerts, while the real criminals were slipping through the cracks. I realized then that the old ways of fighting financial crime were broken. It’s a war we were losing.
Money laundering isn't some abstract financial crime. It’s the engine of drug trafficking, terrorism, and human suffering. We're talking about trillions of dollars a year. And the methods used to catch it are stuck in the past. Rule-based systems are like putting up a chain-link fence and expecting it to stop a tank. The criminals are just too sophisticated. They use shell corporations, synthetic identities, and a web of transactions so complex that it’s impossible for a human to unravel. It’s a big data problem, and you can't solve a big data problem with small data tools.
The Old Way is Broken
For decades, the approach to AML has been reactive. A rule is triggered—say, a transaction over $10,000—and an analyst looks at it. It’s a manual, time-consuming process that generates a mountain of paperwork and very few results. The criminals know the rules. They know how to structure their transactions to stay just below the radar. It’s a cat-and-mouse game where the mice have all the advantages.
I once advised a fintech startup that was trying to get a banking license. The regulators were obsessed with their AML compliance. They wanted to see the rulebook. I told them, “The rulebook is the problem.” A static set of rules is a sitting duck. The financial system is a dynamic, living thing. You need a system that can learn and adapt.
Criminals use a technique called “smurfing” or “structuring.” They’ll take a large sum of cash and break it up into smaller deposits, each below the $10,000 reporting threshold. They’ll use dozens of people, often low-level mules, to make these deposits at different banks and different branches. A rule-based system, looking at each transaction in isolation, sees nothing wrong. It’s only when you zoom out and look at the whole picture that the pattern becomes clear.
AI to the Rescue: Transaction Monitoring on Steroids
This is where AI comes in. Instead of relying on a handful of crude rules, we can use machine learning to look at everything. Every transaction, every account, every customer, all at once. An AI system can see the patterns that a human analyst—or a team of a hundred analysts—would miss.
Think about it like this. A rule-based system is like a security guard with a checklist. An AI is like a detective who has a deep understanding of human behavior. It can spot the subtle tells, the anomalies that don’t quite add up. It can see a series of small, seemingly innocent transactions that, when taken together, form a clear pattern of illicit activity.
We’re using techniques like autoencoders and isolation forests to do this. An autoencoder is a type of neural network that learns to compress data and then reconstruct it. You train it on a massive dataset of normal transactions. Once it’s trained, it’s very good at reconstructing normal transactions, but it struggles with abnormal ones. When it fails to reconstruct a transaction accurately, that’s a red flag. It’s a sign that something is different, something is off.
Isolation forests work differently. They’re a type of ensemble learning method that’s based on decision trees. The algorithm randomly selects a feature and then randomly selects a split value for that feature. It does this over and over again, creating a forest of trees. The idea is that it’s easier to isolate an anomaly than it is to define what’s normal. An abnormal transaction will be easier to separate from the rest of the data, so it will have a shorter path in the tree. It’s a simple but powerful idea.
Connecting the Dots: Entity Resolution
But transaction monitoring is only half the battle. The real challenge is figuring out who is behind the transactions. Criminals are masters of disguise. They hide behind layers of shell companies, trusts, and nominees. It’s a deliberate, calculated effort to obscure the true ownership of assets.
This is where entity resolution comes in. It’s the process of piecing together scattered bits of information to create a single, unified view of a customer or an organization. It’s about connecting the dots. And it’s a perfect job for AI.
I’m an investor in a few companies that are working on this, and the technology is incredible. We’re using graph neural networks (GNNs) to map out the relationships between different entities. A GNN can see that a director of one company is also the shareholder of another, and that both companies are sending money to the same offshore account. It can see the hidden connections that the criminals have worked so hard to conceal.
Imagine a web of companies, all registered in different jurisdictions, with different directors and shareholders. To a human, it looks like a tangled mess. But a GNN can see the underlying structure. It can see that all the companies are ultimately controlled by the same beneficial owner. It’s like having a superpower that lets you see through the corporate veil.
A Real-World Example
I’m an angel investor in a startup I’ll call “Veritas Financial.” They’re using a combination of advanced transaction monitoring and GNN-based entity resolution to help banks fight money laundering. One of their clients, a mid-sized European bank, was struggling with a huge backlog of alerts. They were throwing bodies at the problem, but they were getting nowhere.
Veritas came in and deployed their system. Within a few weeks, they had uncovered a network of shell companies that was being used to launder the proceeds of a massive VAT fraud scheme. The network was spread across a dozen countries and involved hundreds of bank accounts. The old system had seen a few of the individual transactions, but it had completely missed the bigger picture. The AI saw the whole thing.
The GNN was able to link together seemingly unrelated companies by finding common directors, addresses, and even phone numbers in the registration documents. It then overlaid the transaction data, which showed a clear pattern of money flowing from the companies involved in the VAT fraud to a series of shell companies in offshore jurisdictions. It was a classic money laundering scheme, but it was so complex and so well-hidden that it would have been impossible to uncover without AI.
The Human Element
Now, I’m not saying that AI is going to replace human investigators. Far from it. What it’s going to do is make them better at their jobs. It’s going to free them up from the drudgery of chasing down false positives and allow them to focus on what they do best: solving complex financial crimes.
An AI can tell you that something is suspicious, but it can’t tell you why. It can show you the pattern, but it can’t explain the motive. That’s where the human element comes in. You need an experienced investigator to look at the evidence, to connect it to the real world, and to build a case that will stand up in court.
I see a future where AI and human investigators work side-by-side, each playing to their strengths. The AI will sift through the data, find the needles in the haystack, and present them to the human investigator. The investigator will then use their experience and their intuition to make sense of it all. It’s a partnership, a collaboration between man and machine.
The Arms Race
Of course, the criminals aren’t standing still. They’re smart, and they’re well-funded. They’re already starting to use AI to their own advantage, to create more sophisticated schemes and to find new ways to evade detection. This is an arms race, and we need to stay one step ahead.
We’re already seeing evidence of adversarial AI being used to attack AML systems. For example, a criminal could use a generative adversarial network (GAN) to create synthetic transactions that look perfectly normal to an AI system. The GAN would learn the patterns of normal behavior and then generate transactions that mimic those patterns, but are actually part of a money laundering scheme.
That means more investment in research and development. It means more collaboration between the public and private sectors. And it means a fundamental shift in how we think about financial crime. We need to move from a reactive, compliance-driven mindset to a proactive, intelligence-driven one.
We also need to think about explainable AI (XAI). If an AI system is going to be making decisions that have a major impact on people’s lives, we need to be able to understand how it’s making those decisions. We can’t just have a black box that spits out answers. We need to be able to show our work, to explain our reasoning to regulators, to judges, and to the public.
I’m not saying that AI is a silver bullet. You’ll always need smart, dedicated human investigators to make the final call. But AI is the most powerful weapon we have in this fight. It’s the only way we’re going to be able to level the playing field and start winning the war against money laundering.
This isn't just about protecting the financial system. It's about taking away the financial oxygen that fuels some of the worst criminal enterprises in the world. It's a fight worth having, and with AI, it's a fight we can win.
Frequently Asked Questions
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.