Blogs
Financial Crime Is a Network Pattern: Chris Stephenson on Sibos TV
The future of financial crime investigations starts with connected data
Financial institutions are investing heavily in artificial intelligence to strengthen financial crime detection and investigations. Yet one of the biggest obstacles to realising its potential remains the data underneath it.
Transactions, customer records, communications and digital interactions often sit across disconnected systems. Each may reveal part of the story, but understanding financial crime requires investigators to see how those fragments fit together.
At Sibos 2026 in Miami, Chris Stephenson, Head of AI Strategy and Operations at Nuix, joined Sibos TV to explore why data readiness matters, how AI and graph intelligence can help investigators uncover hidden relationships, and what it takes to deliver outcomes that are transparent, auditable and defensible.
His central message: criminal behaviour is a network pattern, and investigations need to reflect that reality.
Watch the full interview below, explore the key takeaways and read the conversation.
Five key takeaways from the conversation
1. AI is only as effective as the data behind it
Fragmented, outdated or poorly understood data can undermine AI outcomes. Before adopting AI, institutions need to understand whether the underlying information is relevant, reliable and ready for its intended use.
2. Financial crime investigations need the whole story
Transactions show what happened, customer and KYC records help establish who was involved, and communications can explain why. Connecting these sources around people, accounts and entities creates a more complete investigative picture.
3. Criminal behaviour is a network pattern
Financial crime often emerges through connections between people, organisations, accounts, transactions and communications. Graph intelligence can help surface relationships that isolated records may obscure.
4. Automation should give investigators their time back
Intelligent automation can reduce repetitive data preparation and workflow tasks, helping investigators focus on relevant risks, relationships and evidence.
5. Defensible AI requires transparency, auditability and human control
In regulated environments, organisations need to explain how outcomes were reached. Chris highlights transparency, audit trails and meaningful human oversight as essential.
Full interview transcript
The following interview has been lightly formatted for readability. Speaker quotations require final video verification.
Why data readiness comes before AI adoption
Sibos TV:
Financial institutions have invested heavily in AI, but do you think the bigger challenge is still the data underneath it?
Chris Stephenson:
Absolutely. You’ve put your finger on it right out of the blocks.
We’ve talked to a lot of institutions this week, and many of them are leaning in heavily to AI. They’ve operationalised it fairly well. Getting the return on that investment is the challenge, and one of the biggest issues there is the data.
Data is the fuel of AI. If you don’t have your data in order, curated and understood, and you haven’t asked whether it’s the relevant data and whether it’s up to date, that undermines the AI outcomes. We’ve seen that run the gamut, from outcomes that are less than stellar, with no return on the investment, to customers coming to us because they had to pull the plug on the initial deployment. Customer data was being exposed, privacy data was unintentionally being exposed, and hallucinations were rampant. It was because the data used in the system was not ready for prime time.
So getting your data AI-ready is really step one, before you even get to AI adoption.
Bringing communications, transactions and KYC data together
Sibos TV:
New data sources such as WhatsApp, collaboration platforms and digital payments data continue to emerge. How can financial crime teams bring that information together with transactions, customer and KYC data to build a more complete investigative picture?
Chris Stephenson:
Financial crime has always been a story. The transaction is the what. That’s the narrative of the story, what happened. The KYC, or the customer, is the who: who the customer claims to be. But the why is in the communications. Right now, the why is sitting inside WhatsApp more and more frequently. It’s sitting in Teams, in Slack, in voice notes. These systems were never designed to hold information that sits right next to the payment record.
So the challenge becomes: how do you get your head around all of that data in a way that centres on the entities involved, rather than the data streams? It’s not a simple challenge, but I think about it in three parts.
First, ingest all the data and normalise it across the various streams, channels and languages. Bring it together into a single view of the investigation, using the entities, the names, organisations, accounts and dates, as the connecting thread. Get your head around the data, normalise it, and use entities as the centre of gravity.
The second is identity resolution, which is really challenging. An individual could be a KYC record, a WhatsApp handle, a phone number, or the recipient of a bank transfer or wire. You’ve got four people who are actually one person, but if you track them as four, you’ve exploded your data and added a lot of noise. Resolving down to that single entity is hard, but it’s an important part of this.
Third is AI, which you can’t get out of any data conversation without mentioning, as much as I try to avoid it because it’s so overused. Use AI as a triage tool, not as a judgement capability. There’s no way humans can effectively get across all the data they need to in order to identify the patterns and relationships. That’s what AI is great for. But balance that with the investigative power of the human, who pressure tests the outcomes, and make sure the explainability comes along to the regulators as well.
Why connecting the dots is still so difficult
Sibos TV:
Why is connecting a person to an account, an entity, a communication or a transaction across huge volumes of data still so difficult?
Chris Stephenson:
Well, it’s buried in the question, because the volume of data makes it challenging right away. Rarely does a criminal show up with a name tag. “Hi, my name is so-and-so, I’m here to rob you.” The same is true of our data. Data is not very well labelled, and unstructured data is, by definition, unlabelled. It doesn’t fit into neat little rows and columns, which makes the problem really challenging out of the gate.
So you’ve got one criminal, 100 fragments, and no one is seeing across all of that to see the single person as a whole.
The idea is that criminal behaviour is a network pattern. If you address it that way, you can see the relationships. The signal is really buried in the connections between the data points and not in any single record, which is kind of mind-blowing when you think about it.
I was thinking about this the other day: volume hides the needle, but fragmentation hides the relationships and the connectivity. It’s a double-edged problem. Frankly, this is one of the reasons we acquired Linkurious, a decision intelligence platform that allows investigators to surface those relationships and avoid spending their time just making queries.
Seeing the whole forest: tables and graphs together
Sibos TV:
What does visualising those relationships differently allow investigators to see that might be missed when information is spread across traditional systems, tables or individual records?
Chris Stephenson:
Tables show records, which is great. They’re facts. Graphs show the relationships and the connections, and the magic is in having both together.
I don’t necessarily need to know how many trees there are in the forest, or even how many kinds of trees. Those are interesting facts. I need to see the whole forest, and I need to know which direction the fire is moving. That connectivity between data points is so critical to solving these crimes, because that piece is usually missing.
Crime is a network pattern, and we should investigate it as such. I don’t hear a lot of people talking about that, although this week I will say the term “knowledge graph” is getting a lot of traction. I think people are waking up to the idea now.
How automation gives investigators their time back
Sibos TV:
With payment fraud and other forms of financial crime becoming increasingly complex, how can automation help investigators move through growing volumes of data faster, identify genuine risk and reduce manual effort?
Chris Stephenson:
This area doesn’t have a data shortage problem. We have a data inundation problem. There’s so much of it.
We’ve seen smart, intelligent automation bring investigative timelines and workflows down from weeks to hours. It does that by automating the process, by eliminating the false positives that send us down rabbit holes we don’t need to go down, and by giving the investigator back the time to go after what really matters. That’s the main driver: giving them that power. And frankly, the ability to see relationships outstrips the typical approach of looking at the knowledge inside a graph on its own.
If you think about it, no investigator ever signed up to be a data wrangler, or to build workflows. Some of them spend hours writing hundreds of steps just to get the data ready. If you can automate that process across those workflows and make it repeatable, think about how much time you’re giving back to the investigator to focus on the real work. That’s the truly inspiring thing: seeing that kind of impact.
Building trust in AI in regulated environments
Sibos TV:
There’s still, understandably, caution around AI in regulated environments. What needs to happen for investigators and regulators to trust an AI-assisted outcome?
Chris Stephenson:
Really, all you need to do is say “trust me”, and then they just do. No, I’m joking. It’s a legitimate problem, especially in regulated environments like this one. Trust is built in three ways: transparency, auditability and human control.
On transparency, you want to be able to open up the black box and show your work. What tools are being used? What models are baked into this? How does it work, and how is it coming up with its answers? If you can shed light on that, people start to trust you as an organisation and trust your technology.
Auditability is key because it’s the evidence of what we did and what the system did along the way. For every workflow and every request: who requested it, what was the outcome, and why did it take that step? Building that audit trail is non-negotiable when it comes to courtroom defensibility, credibility and trust.
The third is human control. It’s great to leverage AI and to know that some things are automated, but having the human in control of the outcomes and decisions is critical to building trust.
The transparency and auditability aren’t just for outsiders to feel comfortable. They empower the human investigator to go back and see what the system did and why, and maybe override it or run it again. It’s not just that the human shows up at the end and says “that’s the right answer” or not. They need to be able to find the proof that got them there, so they can make a strong decision in a defensible way. If you do those three things, you build trust through the evidence.
What changes for the investigator, the institution and the regulator
Sibos TV:
If we manage to get this right, what changes for the investigator, the institution and the regulator?
Chris Stephenson:
I could go off on a tangent on all three, but to keep it short.
For the investigator, we just want to give them their time back. We want to give them their life back. They didn’t sign up to be data wranglers or to build workflows. They signed up to resolve crime. Give them that time back, and their dignity back, to focus on what’s important.
For the institution, we’ve seen the backlogs and how they affect institutions, and budgets are not growing. If anything, they’re shrinking. So you give the institution the ability to get off the back foot, clean up those backlogs, and stay ahead of the problem instead of behind it, instead of being stuck in paperwork.
And ultimately, the regulator just wants what they’ve always wanted: give me the answer, and show me how you got there. And then maybe I’ll get home in time for dinner.
Explore more from Nuix
Turn complex financial crime data into actionable intelligence
Understanding the relationships between people, entities, transactions and communications can help investigators build a more complete picture and make informed, defensible decisions.
Explore nuix financial crime solutions
Continue the conversation
Explore more perspectives on financial crime, AI and connected data from Nuix.