What a Polluted River Taught Napier About Dirty Money

An interview with Dr Janet Bastiman by Sean Murphy

An environmental scientist trying to measure pollution in a river almost never stands at the pipe where the contaminant goes in. They stand somewhere downstream, draw a sample, and read the chemistry of the water as it passes: a rise here, a dissipation there, traces that thin out and then concentrate again as the current carries them along.

It was while reading exactly this kind of research that Janet Bastiman, chief data scientist at Napier AI, had the thought that would become the company's most distinctive piece of detection work. Banks, she realised, sit in much the same position as that scientist on the riverbank. They are very rarely the point at which illicit money enters the financial system. What they see instead is the flow afterwards, the disturbance moving through accounts that are all trying very hard to look ordinary.

That parallel between river chemistry and money laundering is not a marketing flourish bolted on after the fact. It is the literal origin of the method. Bastiman, who holds degrees in molecular biochemistry and mathematics alongside a doctorate in computational neuroscience, has spent more than two decades moving between telecommunications, marketing and financial services, and at Napier she leads the data science behind a compliance platform now used by more than 150 institutions. The river idea came to her late one evening reading papers on detecting antibiotic pollution in waterways, when it struck her that the equations environmental researchers use to trace a contaminant downstream might, with a little data transformation, trace laundered funds in precisely the same way. It turned out they could.

The problem the method attacks is one of the oldest in anti-money laundering, and one of the most stubborn. Criminals obscure the origin of funds by breaking them up, pushing them through many points, and filtering them through accounts engineered to resemble normal customers. Any single bank sees only a fragment of that journey, a subset of a subset, and from inside that narrow window the activity can look unremarkable.

What Bastiman's approach looks for is not the suspicious individual but changes in the amplitude and frequency of transaction patterns across a connected group. Watch a network rather than an account, she argues, and the laundering reveals itself as a kind of collective motion: balances that rise together and then fall together as the money passes through, sometimes quickly, sometimes slowly, sometimes split into different sized pots and spread wide, but still detectable as a single disturbance in the same way a pollutant remains legible even after a river's vortices have scattered it.

It is an elegant change to how detection usually works. Most systems hunt for the anomalous transaction, the outlier that trips a rule. Bastiman is instead reading the wider hydraulics of the network, finding the trigger that simply does not exist when you examine accounts one at a time. On top of that signal sits an AI layer that assesses risk, so that compliance teams are not buried under a thousand indicators of small rises and falls but are shown the pattern where those movements cohere into something worth investigating. That, in a sector drowning in false positives, is the prize.

Much of the work matured inside the Financial Conduct Authority's Supercharged Sandbox, the regulator's flagship innovation programme, which Napier joined for its launch cohort in late 2025 alongside the technology partners NayaOne and Nvidia. Entry was competitive, with the regulator fielding more than 200 applications for a limited number of places, and the format gave participants three months of access to compute, synthetic datasets and expert advice in exchange for a clearly defined proof of concept. Five of Napier's data science team, Bastiman among them, spent the period stress-testing the frequency-based method on far larger and more varied data than the firm had been able to use in house, before presenting the results at the end of January. The internal name was Project Theseus. The output reached the market in March 2026 as Insights AI, a feature within Napier's transaction monitoring product.

What the sandbox really helped with, in Bastiman's account, was not the detection itself but the question of explainability. It is straightforward enough to tell a compliance officer that a given customer and a given set of transactions look suspicious. It is far harder to explain that the suspicious part is happening outside what the bank can see, in the connected accounts and onward flows beyond its own walls, and to write that up in a way that satisfies a regulator and stands behind a decision that may ultimately feed into law enforcement. Resolving how to document network-level suspicion when every existing process is built around the individual account was, she says, the bulk of the conversation with the FCA, and the part that had to be solved before the technique could become a product.

The Napier AI / AML Index for 2025 to 2026 puts the cost of financial crime compliance to UK institutions at more than seven billion pounds in 2024, and estimates that AI-driven approaches could save the country's financial sector around two and a half billion pounds a year. The adversaries, meanwhile, are well funded, run their own data teams, share information freely and are bound by none of the explainability requirements that constrain the institutions chasing them. Criminals, as Bastiman puts it, do not need to explain what they do.

The deeper pull of the river idea is that it changes where a bank should even be looking. No institution can watch every account where dirty money first lands, just as no scientist can stand over every pipe feeding into a river. What a bank can do is study the flow passing through its own accounts and reason its way back to everything it cannot see. Bastiman is honest that some of what comes next for her team, linking up data sources and building proper guardrails around agentic systems, is further than the UK's infrastructure can currently stretch. What she returns to again and again is what the riverbank teaches. The source may be hidden far upstream but the disturbance it leaves behind is not.