How AI can drive productivity and value in the financial sector

High-quality data and a carefully thought-out approach are the essential elements for unlocking the potential of AI.

Artificial intelligence (AI) is booming. Patrice Latinne, Data & Analytics Partner at EY Financial Services, and Nicolas Goosse, Head of Artificial Intelligence at Belfius, discuss the perspectives that AI opens up for the financial services industry.

“Banks and insurance companies have been working with artificial intelligence for several years now,” says Patrice Latinne. “All players in Belgium use AI in one way or another, albeit at different maturity levels. They fit AI into their business strategy and draw up roadmaps that evolve with the technology, their ambitions and their priorities.”

The robotization of processes in back-office functions has paved the way. “The first tools allowed us to improve the ‘cost to serve’ of our operational functions,” says Nicolas Goosse. In the beginning, it was about automated algorithmic models that allowed for a certain amount of prediction. At that time, there was no advanced artificial intelligence that enabled machines to derive rules automatically and learn by themselves.
 

Fighting fraud effectively

AI tools really made their appearance in risk assessment. They prevent an insurer from overpaying compensation or help a bank to anticipate defaults on loans. AI tools also play a vital role in the fight against fraud and financial crime. “Banks operate within a very strict regulatory framework,” says Nicolas Goosse. “But the creativity and resources of fraudsters are constantly evolving. Thanks to AI models, you can detect abnormal behavior and fight fraud more effectively.”

The models that serve to refine 'customer risks' are not only based on financial ratios and the intrinsic characteristics of abnormal transactions, Patrice Latinne believes. Large amounts of external public or paid data are automatically analyzed, aggregated and integrated by AI. It contains information from financial media but also an increasing variety of market data.

Responsibility and inherent risks

New technology in a bank or an insurance company also entails risks. Information automatically produced by AI can, for example, have consequences for a quotation, a communication or the processing of a customer file. If that information is incorrect or imprecise, it can harm the customer, with all the implications for the reputation of the company.

“The data on which machine learning models are based can be biased,” says Nicolas Goosse. That is why Belfius has drawn up a very strict framework for the use of AI, with an eye for the ethical component. Meanwhile, the R&D teams do continue to explore new areas. “We use AI mainly as a copilot. There is still an employee between the customer and the machine, to control that machine and avoid any risk for our customers.”

Patrice Latinne points out the risks inherent in the development of the solutions. “The human factor is decisive. You have to ask the right questions. Have the data scientists and developers properly integrated all the dimensions of a problem when building the models? Are those dimensions properly reflected in the code that ChatGPT generates based on textual instructions? Are the data sources clean and well-managed? With the recent development of generative AI, will a consumer - employee or customer - still be able to discern that a piece of content was generated entirely by AI? Are all relevant departments in an organization, including compliance officers, sufficiently involved? All these questions emphasize the responsibility of every organization offering a product or service with artificial intelligence, now and in the future.”

European fundamental rights

The European Union is therefore working on a law around artificial intelligence (AI law). The regulation, due to enter into force in 2025, will ensure that AI systems placed on the European market are safe and respect citizens' fundamental rights and EU values. But doesn't the rapid rise of generative AI make such an attempt at legal regulation futile? “The logic of the AI ​​law remains fully relevant to generative AI,” says Patrice Latinne. “But when you see how quickly generative AI is developing, banks and insurers are mainly wondering if that legal framework will arrive on time. Will it be current enough? And what about the competitive pressure to deploy such technology quickly?

The stakes are high, as it involves the management of highly complex, yet easy-to-use systems with billions of parameters.

Patrice Latinne concludes: “Today, when we think of AI and banks, we think of automating tasks, such as filling in Excel tables, reading and summarizing documents, and automatically adjusting clauses in a contract. But tomorrow, AI-powered solutions can make an enterprise's management systems interactive. This way, they can interact with each other and with the employees. And that increases productivity and generates added value. So both data quality and governance, and control processes are crucial.”