In the year since the 2017 Banking Growth Forum, Artificial Intelligence (AI) has continued to advance. Algorithms have beaten the best human players at increasingly complex games, including Go, poker, and Dota 2. Machines have proven they can teach themselves new skills such as bluffing, learn from results to play better and better, and even innovate new winning strategies.
But can AI answer a question like “What rate should I charge for a particular product to maximize contribution without causing unacceptable loss of new bookings or an increase in risk?” Answering this kind of question requires understanding causality as well as correlation, and the magnitude of the underlying relationships is as critical in financial services as it is in healthcare.
At the 2018 Banking Growth Forum, I’ll talk about how close artificial intelligence is in 2018 to answering hard questions about how customers will respond to different offers at different prices and how this intelligence can be used to target the right product at the right time to the right customer. And I’ll look at other current and potential impacts of game-winning technologies as they move increasingly into the financial services.
Artificial Intelligence and machine learning is predicated upon the fact that we have much more information about individuals than we had in the past – by a long shot. Say we have information on ten million customers and prospective customers. In the past, we might have a thousand pieces of information about each customer – for example, the entries in a credit file plus records of transactions with our bank. Traditional statistical approaches could pick out which among these pieces of information were most predictive of price-sensitivity or probability of default. In the new world of “big data” we may have 100’s of millions of pieces of data associated with each customer such as social network activity. AI algorithms such as deep learning are able to “mine” this much richer data set to find subtle relationships that are far more predictive than traditional approaches. The relationships can be used to improve product targeting, underwriting, and pricing decisions.
However, this does not mean that applying AI to financial services is not without its challenges. Artificial Intelligence algorithms learned how to win at Go by playing the game against themselves a billion times. However, financial service providers are much more limited in the extent to which they can experiment with customers. Furthermore, understanding causality is key – Google does not need to know “why” or even “how” the AlphaGo program consistently beats the best human players in the world – it only matters that it does. But, when it comes to customer-facing decisions on underwriting, product selection and pricing, financial services companies, regulators and customers are likely to be asking both “why” a particular decision was made and what actions they might take to change it.
Furthermore, AI not only has to contribute to predicting a behavior or prescribing a treatment, but also to delivering fair decisions and a outcomes that work for both banks and customers. I have a lot to say about that—and I want to hear what you think.