By Rob Cutler, Managing Director, NexusAML.
AI is changing financial crime operations quickly. Routine screening, data analysis and case preparation can increasingly be automated, allowing analysts to focus their time on work that requires more investigation and judgement.
That shift should improve efficiency. But it also raises a less obvious question. If AI takes on more of the straightforward work traditionally handled by junior analysts, where will the next generation of experienced investigators build their judgement?
Nexus AML recently surveyed 100 senior FinCrime professionals about analyst development, training and the impact of automation. The findings suggest this is becoming a genuine concern.
Historically, analysts have developed much of their operational judgement through an informal apprenticeship. They learn the policies and procedures, start with relatively simple cases, see the same patterns repeatedly, encounter exceptions and gradually progress towards more complex investigations.
The problem is that this pathway is becoming less dependable.
71% of respondents said junior analysts do not receive sufficiently broad exposure to different customers, structures and risks. Meanwhile, 81% were concerned about a future shortage of experienced operational judgement if development opportunities continue to narrow.
Automation could accelerate that problem.
Routine work is not necessarily the most valuable work an analyst will ever perform, but repetition has played an important role in development. It helps analysts recognise common patterns, understand what normal activity looks like and develop the confidence to recognise when something does not fit.
If technology removes a growing proportion of that work, firms should not preserve inefficient manual processes simply for training purposes. They do, however, need to consider how the experience previously gained through those processes will be replaced.
Knowing the rules is not enoughÂ
This matters because financial crime decisions rarely depend on identifying a single red flag.
Analysts may need to consider transaction activity, customer behaviour, ownership structures, source of funds, jurisdictional exposure, adverse media, sanctions or PEP exposure and the quality of the available evidence.
Policy provides the framework, but judgement determines how those different pieces of information should be interpreted together.
Our survey found that 78% of respondents believe existing training does not provide enough exposure to realistic decision-making. This helps explain why traditional classroom learning and computer-based training can only go so far.
Knowing a policy is different from applying it confidently when evidence is incomplete, several explanations are possible and the correct next step is unclear.
That is where simulation can play a different role.
Creating experience deliberatelyÂ
65% of respondents selected simulation-based training as the approach with the greatest potential to improve analyst development, compared with 21% selecting AI and automation and just 7% choosing current training approaches or existing CBTs.
Simulation allows firms to recreate the decisions analysts will eventually need to make in production, without placing a real customer or investigation at risk.
Instead of answering a series of knowledge questions, an analyst can work through a realistic case. They can review customer information, investigate transactions, identify missing evidence, apply policy, decide whether escalation is required and record the rationale behind their conclusion.
Crucially, firms can control the experience the analyst receives.
Common situations can be used to build confidence. Variations and exceptions can test whether the analyst really understands the underlying decision. Rare but important scenarios can be practised deliberately rather than waiting months or years for them to appear in a live queue.
Simulation can also test more than whether someone reached the right answer. It can assess the quality of the investigation, the evidence considered, the rationale given, escalation decisions and whether the resulting action was appropriate.
That creates a much clearer picture of whether someone is genuinely ready for greater responsibility.
Preparing analysts to challenge AIÂ
There is another reason this matters.
As AI becomes more involved in financial crime operations, analysts will increasingly need to assess machine-generated recommendations rather than simply perform every stage of an investigation themselves.
That requires judgement.
An automated system may identify a pattern, summarise a case or recommend an outcome. The analyst still needs to understand the context, identify missing evidence and recognise when the recommendation does not stand up to scrutiny.
Training therefore needs to prepare analysts not just to use AI, but to challenge it.
Simulation can deliberately present analysts with both sound and flawed automated recommendations, forcing them to decide whether the evidence actually supports the suggested outcome.
The objective is not to slow automation down. It is to ensure that efficiency today does not create a capability shortage tomorrow.
As routine work changes, firms have an opportunity to move from experience gained largely by chance to experience built deliberately.
The organisations that do that well will not need to choose between automation and human judgement. They will use one to strengthen the other.
For a deeper look at how simulation-based training can build real-world financial crime judgement, read the Nexus AML paper HERE.
Subscribe to our newsletter
