2.5: Beyond good ideas – Making AI work in production for surveillance

Innovation is never easy – and delivering it to production is even harder. Banks should keep this front of mind when they’re shopping for surveillance tech partners.

Choosing a vendor for any next-generation technology involves a delicate balancing act, with banks and brokers needing to temper their hunger for the best innovation with a degree of pragmatism. In the case of surveillance tools, they must factor in not only the features on offer but the firms’ track record at delivering innovations from the lab into production and for ensuring the solution satisfies regulatory requirements.

“For surveillance innovations using AI or machine learning, the migration from proof of concept to production can bring huge benefits for banks, but also challenges,” says Tim Estes, president and founder of Digital Reasoning, a provider of AI-enabled conduct surveillance solutions.

“While a new technology’s proof of value can be very exciting, there are always risks. Unforeseen regulatory requirements have to be reacted to and it’s important to have full transparency about a solution’s ultimate capabilities,” he notes. “Senior leaders should be mindful of both.”

Case studies: Two paths
The following real-world examples highlight some of those risks, but also ways in which banks and brokers with time and resources to invest can overcome them through closer collaboration, Estes says.

In case study 1, a broker dealer ended up being fined, partly because of its use of conduct surveillance tech supplied by a less experienced vendor, as well as insufficient focus on ensuring the solution was regulation-ready. 

In this instance, the vendor was selected through a head-to-head proof of concept competition with a rival that saw both tasked with finding emails that were seeded with conduct-breach language that had previously been used and was memorisable. It took the selected vendor – a start-up that was proving out new technology using behavioural analytics to reduce false positives – 12 months to reach production delivery. The broker dealer that implemented the solution, however, received a regulator query that it was unable to respond to, contributing to it being fined later on.

In case study 2, close collaboration between a global European bank and its chosen surveillance tech vendor eventually led to success. 

The two worked closely on a proof of value together, using natural language processing (NLP) to reduce false positives and eliminate language patterns that had created multiple poor hits and wasted enormous amounts of human time in review. The vendor also cooperated with the client to make appropriate interfaces and user components to meet regulatory audit. 

It took 18 months for them to achieve proof of concept, after which the bank became the first to deploy NLP-based surveillance at a geography at scale, before going on to roll it out globally. “This was time consuming but ultimately a success story,” Estes says.

Lessons learnt
A number of lessons can be drawn from these two case studies.

First, proof of concepts that don’t consider the risk of implementation and of not receiving regulatory sign-off may ultimately be counter-productive, Estes argues.

Secondly, if one vendor has a greater history of delivering solutions to production but offers slightly less than its nearest competitor in terms of features, it might be worth exploring with them whether extensions or key customisations could bridge that gap, Estes advises. “Is their solution something that could be added on to through point technology or could you ask them to make changes in order to secure the contract?” he asks.

Alternatively, if a vendor does not have a delivery track record but offers truly compelling technology, that too can work – but only if a bank is prepared to invest 12 to 24 months and support them on the journey to production success with technical and subject matter expertise. 

“The idea that you can take an unproven vendor plus have them handle it all is setting yourself up for massive risk,” Estes says. “You have to be prepared to partner with that vendor closely and invest a lot of resources to see that through.”

In it together
Digital Reasoning has itself benefited from this kind of support and, Estes says, remains “deeply thankful for the two or three large names that were willing to invest in us for one or two years to get through that journey together.”

Ultimately, banks and brokers need to be confident that the surveillance tech partner they choose is equally committed to the long haul, he argues. 

“You have to know the other guy’s going to be around and that they care as much about servicing clients they’ve already won as they do about winning the next one,” he concludes. “In this very tight-knit market, a vendor that cannot deliver or provide the appropriate regulatory support is not going to be around for long.”

Based on an international benchmarking survey collecting the views of industry leading experts from 15 of the largest financial institutions globally, the 2020 Surveillance Benchmark Report provides a unique insight into the maturity and development of surveillance functions over the last 12 months, as well as predictions for the future. Including in-depth commentary from regulators, practitioners, consultants and technology experts, it is the only report for professionals in the industry.

Lead sponsor


Partner sponsors



Eventus Systems logo 1


Researched and published by