I was invited to join the panel of the ACAMS ‘Connecting the Dots’ event last week in London. The session explored how AI & Data Analytics are helping to combat financial crime and compliance in financial services.
The conversation was fascinating with contributions from a panel also including senior executives from Barclays, Deutsche Bank and BNP Paribas alongside experts from Exiger.
We discussed a number of different AI related topics and this post shares some of the many highlights from an excellent evening.
AI that explains decisions
One topic that gained particular focus was the requirement for machines to explain themselves. In an industry where decisions carry high risk and high consequence, machines that make complex judgements are a necessity and not just a nice to have.
The panel discussed the necessities of explanations that are human readable as well as discussing how teams that manage AI based machines can benefit from transparent, assured and auditable output.
Expert modelling was regularly referred to throughout the evening with many of the panellists advocating the involvement of in-house and industry experts to identify needs and define models. One of the many benefits is that machines then begin to think and explain in the way that humans need them to.
At Caspian we base machine modelling on gold standard analyst expertise for many reasons, including the way that decisions are explained. If you’re interested in reading more on explainable AI technology, then this article goes into greater detail.
Integration into organisations
We discussed the value of integrating technology-based solutions into existing organisation processes and infrastructures. This is never an easy thing to do given the size and complexity of some financial services firms alongside the disparate legacy nature of embedded systems. Implementation solutions including API, cloud and secure data sharing are enabling real time benefits and we talked about the benefits to two particular functions:
- QA – teams get readable insights to optimise accuracy and avoid bias as well as creating transparent audit trails that meet internal and authority standards
- Analysts – teams get insight that both support development of analysts and encourages their feedback to improve machine effectiveness (which in turn augments superior team performance through detailed risk management)
The potential for Open Source technologies was also recognised through several panel questions. Not only are low costs a major benefit to accelerating development and implementation but the open nature of some of the excellent tools provides significant speed, agility and flexibility when it comes to integration across technology platforms.
Data drives AI success
Several viewpoints were shared as to the importance and powerful nature of data in relation to the success of AI and analytics. David Ng Fat from Deutsche Bank recognised the need for any organisation holding significant quantities of data to organise data lakes as a first step. This sounds a lot easier than it actually is when data is held disparately across systems or labelled inconsistently. If AI models are to be trained or reliant on specific traits of an organisations own data, then this is a critical process.
There were also some excellent examples shared as to how data is used in analytics to uncover previously unseen patterns. Paul Horlick from Barclays referenced some relatively simple yet powerful applications that had helped them to track individuals likely behaviour via propensity modelling. By using expert insights, they were able to find patterns in data that predicted behaviours based on historical data.
AI and the future of humans
The misnomer that humans will be surplus in the future is one that was tackled with an open belief that there is definitely a future for the human brain alongside technology developments. Two particular areas were discussed:
- Training the machine – humans are vital to training machines to think as they do by inputting expert best practice that can explain outputs in a readable format. Teams that manage a machine solution need to define mechanics including risk tolerance levels as well as defining new typologies that mathematics may uncover
- Augment performance – many high volume or complex tasks benefit from the consistent processing power of machines. This is definitely the case in risk investigations where missing areas of information and answers that analysts need can be turned around rapidly. This frees them up for higher value investigation activities that machines recommend an analyst should undertake.
One piece of advice
Asked what one piece of advice we would give to the audience to help them start and get comfortable with AI or analytics, I shared the following:
Don’t start with the technology. Start with your business requirements and not the maths or tech that someone is trying to persuade you is the best in the world. Be clear on your objectives and select a solution that delivers to your needs. Your own business and risk levels will have a big influence on your needs. I look forward to a day when we don’t talk about AI or Machine Learning as being the solution, rather we talk about solving financial crime in the most effective way using technology to assist.
Thanks again to ACAMS and Exiger for welcoming Caspian to the event. This kind of discussion and awareness is vital to help realise the opportunities we have in front of us all to better tackle financial crime and compliance using AI and technology.