European bank uncovers hidden financial crime connections with Neo4j

Neo4j works with many banks around the world. One of the longest established European banks serves millions of clients across Europe, operating for hundreds of years, with the intention to operate for hundreds more. With this perspective, ensuring compliance with regulatory requirements is a top priority. Anti-financial crime work isn’t managed as a legal cost to minimize; it is treated as part of compliance and risk management that will help keep the bank and its customers secure.

At the end of 2019, the bank ran a strategy review of its compliance function and found what most European universal banks had at the time: solid legal processes and established tools, but no data-driven view of financial crime. The bank’s board set a target to change that. The bank’s compliance analytics leader described the ambition behind it: “We want the compliance division to become a data-driven function on a solid legal background, because that’s really the value we can bring to the financial system, by making sure that our part of the financial system is clean.”

Deploying machine learning 

The bank built its analytics capability in-house initially. As machine learning models expanded into anti-money laundering, detection improved. But the models worked from data with a structural gap: they could score a transaction or a customer, but they had no systematic way to represent who was connected to whom, and this was the crucial gap for existing anti-financial crime programs.

“What we were really missing was network information,” the leader said. Money laundering and sanctions evasion routinely rely on parties that look unremarkable on their own and only look suspicious once you can see how they connect to each other, as they transfer money from one shell company to another.

Choosing a graph database

The bank ran a formal evaluation of graph database vendors in 2023 with an outside consulting firm. Neo4j won on four grounds:

1. Durability. “It’s important that we work with trusted counterparts that are also going to be around in the future,” executives said. For an established financial institution, choosing a market leader carries value.
2. Running graph algorithms at production scale. Rather than simply store connected data, the bank’s engineering team wanted to apply Neo4j Graph Data Science to run embeddings and graph algorithms at scale.
3. Visualization. The compliance leader had seen other institutions pair a graph database with a weak front-end and watched adoption suffer as a result. “Ultimately the investigators are the end-users, and they need to be happy,” he said. Neo4j Bloom’s native integration with the graph gave the bank’s investigators a visualization layer the bank didn’t have to design or maintain itself.
4. An actual graph database versus a graph-shaped feature. “There are vendors out there that give you a package that is not really graph analytics, it looks like a graph, but doesn’t have the flexibility of a database,” he said. The bank wanted a graph platform the company could build multiple compliance use cases on top of, not a single packaged tool.

“Graph analytics and graph databases are a super relevant pillar for data-driven compliance and analytics functions.”

The demo that changed the conversation

During an early proof-of-concept session, the Neo4j team ran entity resolution against the bank’s own data. Leaders still discuss this story when showing the tool internally. “The data is there, but the connections were hidden — and the graph database easily shows these connections,” he said. Relationships between counterparties that had been sitting, unseen, inside the bank’s existing records now appeared visually connected on screen, easy for people to understand. That demonstration convinced senior management the technology could add value, well before a production system existed.

Through 2025, the bank scaled network analytics use cases, followed by a technical review that included a GraphRAG workshop. Today, it runs Neo4j Enterprise Edition on-premises, with a graph built from transaction data, customer records, along with account data. Leaders describe it as a connected clone of the bank’s compliance data, built on a data model other teams can extend. Analysts across the bank’s anti-financial crime teams use Neo4j Bloom daily, and other business units inside the bank have begun building on the graph.

Creating a more complete picture

Analysts work through more information than before, and find connected details and context as a result. Analysts don’t run a full network analysis on every alert; they reserve the technique for cases where the added depth is worth the additional time. The reasoning is the same one executives use to explain why transaction lists never worked for this kind of investigation in the first place: “No human can really effectively read something out of 10,000 payments in a list. That’s just not possible. But if you give someone a network of a client, people immediately understand what’s going on.” 

What changed is how much of the data investigators can actually use with the additional context of network connections. “It’s the same data,” he said, “but it’s just much more information that you get out of it.” 

“No human can really effectively read something out of 10,000 payments in a list. That’s just not possible. But if you give someone a network of a client, people immediately understand what’s going on.”

What’s Next

The bank is now feeding graph-derived features into its machine learning models, starting with fraud detection and moving to anti-money laundering next. A second work-stream, still early, uses large language models to pull relevant graph context into an investigator’s case file alongside other information. The compliance lead is deliberate about where each technology belongs. “For the detection and the analytics, it’s about machine learning,” he said. Generative AI’s role, in his view, is to make investigations more efficient once a case is already open, not to replace the detection work itself.

The bank’s security requirements for the deployment were demanding, and Neo4j adapted the product to meet them, a dynamic executives credit to the company’s size and its responsiveness moving together rather than trading off against each other. “Neo4j is an established company, but you still have the agility of a startup,” he said.

For the bank, the work connects back to the horizon that guides compliance functions: a bank that has operated in the region for generations and intends to continue doing so for many generations to come. “Graph analytics and graph databases are a super relevant pillar for data-driven compliance and analytics functions,” he said. Investigators now see connections that were previously difficult to identify in the bank’s existing data. That’s the difference the bank set out to create in 2019, and is building on every day.

“Neo4j is an established company, but you still have the agility of a startup.”