“Neither RedBox nor Netflix are even on the radar screen in terms of competition.” —Blockbuster CEO Jim Keyes, speaking to investors in 2008

Source:The Book Bank 4.0 by Brett King

Commercial banking can be broadly broken into retail (serving end consumers) and wholesale (serving companies) banking. At a high level, the commercial banking business model may seem simple. Commercial banks take deposits from savers and lend them to borrowers by charging interest. This business model’s core capability is assessing the loan riskiness extended to borrowers. Failures to manage this risk have led to many instances where banks have collapsed (and this is not news to anyone).

AI can help manage this risk better. But AI can help banks streamline many of their business processes beyond risk management. An analysis by Forbes in 2018 suggested that the banking Industry can save
more than $1 trillion by leveraging AI prudently. One AI algorithm that can be very useful in commercial banking is reinforcement learning.

Let us review an example. As mentioned previously, assessing borrower risk is one of the core capabilities of commercial banking. In simple terms, two points need to be analyzed: The risk profile of the borrower and the loan funding costs. Reinforcement learning can help address risks of the loan funding processes. Let us understand how.

Three entities are involved in the loan funding decision process- the bank, the depositor, and the borrower. In machine learning parlance, these are called agents. As you can assume, for any bank, there are thousands (millions in some cases) of depositor and borrower agents.

In the context of this problem, the bank’s objective is to generate a loan and deposit pricing grid. Simultaneously, the bank also needs to calculate its profit/loss and self-funding status at any time.

A reinforcement learning algorithm can be trained to update the pricing grid while incorporating feedback from recent actions’ impact on the profit, loss, and asset and liability ratios.

The algorithm must account for certain behavioral aspects of the two key agents, the depositor and the borrower. For example, on the deposit’s maturity date, the model will estimate depositor behavior on whether they want to keep or withdraw the deposit. The algorithm will simulate the depositor’s decision by randomizing the decision as a % chance of expected interest.

While this is just one example, reinforcement learning can also be leveraged in other commercial banking areas. A reinforcement learning model designed for funding decisions can also be repurposed for different regions within retail banking. Opportunities to transform commercial banking with AI algorithms are
plenty.

References:

• Bank 4.0 Author: Brett King
• AI and the future of banking. Author: Tony Boobier
• Artificial Intelligence in Finance. Author: Yves Hilpisch