Can generative AI help tackle anti-money laundering? Yes, says WorkFusion

The banking and financial services industry is not very keen on taking risks. Take generative AI for example. As the technology progresses it can come up with the right answers a surprisingly large amount of the time, but not at a high enough percentage for financial institutions working on, for example, anti-money laundering (AML) to risk using it. 

As Peter Cousins, Chief Technology Officer of digital workforce provider WorkFusion, puts it: 

Our customers can’t afford a single mistake. Generative AI is powerful, but is too slow, too expensive and has too many random behaviors to be used in a regulatory environment entirely on its own. But when it is supported by people and other AI approaches you can get great results. Financial institutions can leverage the benefits of gen AI without the added risks.

WorkFusion is tackling concerns about risk from the financial services market by adding generative AI enhancements to its existing pre-built AI digital workers, and stakes the claim that by putting the two together the risk of errors has in some cases come down to nearly zero, whilst automation rates for decision making can rise by up to 95%.  Cousins adds: 

Using gen AI means we can take another crack at a decision. A digital worker could say, ‘I’m pretty sure this is a false positive’. And normally when they’re not 100% sure it gets routed to a person to adjudicate. That undermines the value proposition of the software because it’s supposed to be trying to take as much of this workload off people as possible.

When we use gen AI to make a decision, we have found that when the gen AI and the model that is not completely confident agree, it’s right virtually 100% of the time. Only then if there’s any doubt, it is sent to a person, so the need to involve a person in the process decreases significantly.

Adding gen AI to WorkFusion’s digital worker model also gives its customers a risk free, more effective, faster way of handling the vast swathes of non-standardized documentation thrown up by AML and other financial crime investigations, he argues/ 

The pre-built models already handle the most common and important document types, but for other, non standard documents from different geographies and regulatory regimes, they have to rely on training these document extractors, which can often be a huge effort.  Cousins says: 

Just because you have a bank statement extraction model, doesn’t mean it’s going to be able to handle every single bank statement available.”

There can be a long trail of bank statements and other documents, and the more variations there are, the more training the model requires, he adds: 

Now with LLM in the loop, the gen AI will get the first crack at it before the person. We can run that through processing, and if it was a complete failure the gen AI can do the document extraction and then a person can just approve it.”

No Code

WorkFusion has put a lot of energy into incorporating gen AI into its no code digital worker platform, according to Cousins.

Most of the behaviors of the digital worker can be significantly changed and enhanced by using a UI that will control their behavior. There’s a UI-based approach for getting data from virtually any legacy systems, databases, public data sources, third party apps, without having to do any programming.

There are ways in which we can massage the data once we get it, reshape it, split it, join it, but we can also do things like compute metrics that are important for looking at a case. These can be done without writing any code. You can write business rules that say, ‘If it’s a payment, that’s a real-time payment, which is non-recallable, and if it’s over a million dollars and it’s going anywhere to a country that borders Russia, I immediately want it escalated to an analyst for review’. That kind of rule can be entered into the system as a UI and No Code is fundamental to how we make the digital workers perform at a higher level. 

The company is providing gen AI Chat interfaces to these features, so the business user can describe what they want to see. It turns it into the structured form in the UI on the screen, that can be modified until it’s correct. Cousins explains: 

We’re doing this for all of the No Code features, transformation, the rules, ultimately everything. And this is important because it really makes more practical the idea that the domain experts in anti-financial crime will be able to run experiments entirely on their own without having to involve IT, and they may still have change control, that is maybe in partnership with IT. There can be guardrails to make sure that proper regression testing was done, before it goes live into a production environment, but it’s a very powerful tool. And we found that with the digital workers, the more the domain experts are able to enhance them, the better they work, and that winds up being different in each financial firm.

Adding gen AI to its digital worker platforms has also enabled WorkFusion to offer banks and financial services customers more effective ways of producing detailed narratives of any regulatory investigation. The firm pitches that it has reduced the time it takes to summarize and analyze the documentation, and speeded up the time it takes for analysts to make a decision about any unusual activity. This means that AML analysts are now able to collate data from different sources and simply review it.

Case notes can be generated using gen AI to explain why the problem was not a problem, or why it is. Then analysts can decide whether to escalate the issue, or file a suspicious activity report.  Cousins says:

We’ve been using gen AI for doing this last stage of the investigation, and the dossier work. And that involves two things. What we found in AML, for example, is that 60 to 80% of time spent is on gathering and massaging the data. But at the end, you want to be able to create a synopsis of all the data that was reviewed. 

Customers want to be able to create a stack and rank of the most material factors that they should look at. They want to be able to summarize those as a paragraph that can be entered into the system of record where the case was being maintained.  Being able to do that with AI, and being able to have a Q&A interface where the investigator can ask a question of the data, speeds up the time for an analyst to make a decision, so they can quickly file a suspicious activity report or make a decision about an alert.

My take

The fInancial services industry wants the benefits of using gen AI, whilst avoiding any of the risks of using a fast developing technology. WorkFusion’s two tier approach of blending gen AI with its existing AI enabled digital workers seems to promise those benefits with some impressive productivity improvements, whilst still avoiding risk, and supporting regulatory compliance.