Dr. Edwin Van der Ouderaa is an Advisor to Financial Services boards. He has a Ph.D. in Information Theory, Electrical and Electronics Engineering. He is based in Belgium.
I was one of the founders of Accenture’s Applied Intel practice and led AI for Financial Services for years. I also ran FS Digital at Accenture and Customer Sales & Service (CRM, call centres, front-offices, etc.) at Accenture Song. My practice totalled $7bn or 10% of Accenture’s revenue. I have a PhD in Information Theory and was privileged to get classes from Marvin Minsky, founder of MIT’s AI lab as my alma mater in Brussels was the first to have an AI lab in Europe, a satellite of MIT.
My work at Accenture allowed me to pioneer Analytics, Big Data, ML and AI, especially in FS, from the front to the back office. My number crunching of the historic data of the top 3000 banks of the world, and the models on their balance sheets that followed from that, got me to be invited to be part of the project team at the Basel Committee, that created Basel III. In fact, I was asked to chair the Liquidity kick-off sessions.
Over the last ten years we pioneered a lot of deep Machine Learning applications in Banking and Insurance. They range from customer-related topics like propensity, pricing, KYC to underwriting and augmented credit scoring over AML and fraud to capital, liquidity and risk models. On the customer side we also worked on real-time personalisation and chatbots.
These models proved to be very powerful and rapidly became a mainstay across most banks. The wave of Digital transformations of the last ten years in Financial Services is in essence the result of this technology in combination with the advent of smartphones with 4G.
Powerful as ‘classic ML’ is, we were also acutely aware of its limitations. That is precisely why I believe that Gen AI can take things to a whole new level. Large language models allow us to do a whole set of things we could not do before and that represent a step change.
It starts with us being able to communicate with the system in normal language. This allows a richer instruction set for systems that often need to cater for very nuanced and complicated situations. One of the Achilles heels of existing decision systems is that it is very hard to code how the system needs to react to all the possible variations in input.
As a result, rather than building complicated decision trees, that typically call ML Python routines, and that always result in complex hard-wired decision paths, which invariably become too hard to understand and maintain, with LLM models we can have the system make cognitive-like decisions on real-life situations. We can throw out the chat bots. We can throw out the RPA, or reduce it to a simple API-layer to connect to the legacy systems. We can have sophisticated personalisation, underwriting, credit scoring, better predictions of potential deteriorating credit, etc. We can have these systems process millions of “unstructured” documents, video and sound recordings, cross-reference and catalogue them. KYC, KYB, AML and fraud can now be handled much more powerfully as we can finally combine structured and unstructured data the way we need to.
A good example is claims handling. I used to need a team of 30 people, a mix of consultants and the top claims experts of the insurance company, to examine over 10 weeks or so, a set of closed claims files. By deeply analysing these historic claims adjustments, we could determine what is called the “leakage”: the difference between what was paid out and the theoretical optimum. By improving the processes, training and decision structures, we could typically reduce up to 40% of the leakage, a massive impact on the combined ratio and bottom-line. But the drawback was that you would need to redo this exercise every few years. We also had to leave a lot of money on the table.
Now with Gen AI, we can have an AI claims expert analysing all cases continuously, in real-time, reducing likely leakage with up to 70%, about double what we can do today, and keep this up. And get better at it, all the time. We can now also automate a much larger set of the simpler claims and push them through straight-through-processing.
Another example is a recent digital bank my team and I built, one of the leading ones in the world. It broke ground with how it allowed real-time consumer credit, inside a wide partner ecosystem, with its sophisticated deep ML models, regulator-approved. We were able to get up to 98% of credit applications in STP, while going down about 100 points in Fico score, without any deterioration in NPLs. Better finding the good apples between the potentially bad ones as it were. But now, with Gen AI, I think we can do even much better than that, going deeper into the Fico scores, get more accurate technical and commercial pricing, and increase the amount people can safely lend.
These examples illustrate how in FS, Gen AI has a double impact: one on the ”technical” content of the business like credit risk or claims adjustment or exposure risk; the other one on further automation of processes, making STP even more robust. I can see the combination of orchestration engines like ServiceNow or Pega with LLMs as the new backbone of front-to-back processes, in a way we could not do before.
An additional benefit of Gen AI based business architectures in FS is that we can have much more flexibility in products and services that can be offered to customers – private banking and a personal broker for all as it were. Processes will be able to dynamically reconfigure depending on customer needs.
In my white paper I also spend a lot of time on explainability and controllability of these AI models, essential capabilities to make the whole thing work well and keep the regulators happy.
To conclude, I would predict the possibility for an improvement of 5 to 10%points to the cost-income ratio in Banking and to the combined ratio in P&C insurance. I also predict we can have 1 to 3% in ROE improvement through better risk, portfolio, capital and liquidity management. The next ten years will be very exciting in Financial Services.
Please check out my White Paper called “2024 – The AI revolution has begun” for a lot more on Gen AI. You can find it on LinkedIn:
Read more:
Personalization also Cultivates Loyalty & Trust
CEO Panel: Synergizing Upsurge & Risk – Mastering the Growth Continuum
