If 2019 was the year artificial intelligence (AI) went from idea to reality for many lenders, 2020 is going to be the year AI becomes a must-have. A November survey of 1,500 top executives by Accenture found that 72% of banking heads say their companies risk going out of business in five years if they don’t scale AI. An even greater percentage say they won’t achieve their growth objectives without more AI. So as financial institutions hit the gas pedal on AI tools like machine learning (ML) here are four trends to watch for in the coming year:
Explainability takes center stage. ML is appealing to lenders because it can expand the borrower pool while reducing charge offs. But in financial services, the journey to that loan approval is a critical part of the process. To really understand your system and defend it to regulators, you have to be able to pinpoint how ML’s complex math connects the dots. Otherwise, you run the risk of producing biased outcomes and worse, you can’t even explain where they came from. That means truly transparent ML is going to be a crucial edge for those who have it and a roadblock for those who don’t.
This idea is spreading. Google recently announced that it is rolling out free ‘Explainable AI’ tools for Google Cloud users. While that’s a good start, lenders need to make sure their explainability is thorough enough for regulators who may come calling and that it can work across cloud platforms and even with on-premise servers. Banks should make sure they are working with partners who truly understand lending.
More data becomes more accessible. This year, there will be about 1.7 megabytes of data created for every person on Earth every second. Much of that — think college football chat boards and cat memes — obviously won’t be useful for financial services, but there’s still a tremendous amount of potentially useful information out there, such as consumer-permissioned checking accounts and tradelines such as utility and cell phone bills. In some of the models Zest AI has built with our lending clients, this kind of credit-adjacent alternative data has provided more than 20% of the performance improvement over traditional underwriting models. Sorting the wheat from the chaff is the difficult part.
Automated AI systems allow companies to ingest more data than ever and well-designed ML models can help credit underwriters use more of the data they already have. The best search engines (including those devised by Netflix and Amazon) use ML algorithms to let the data figure out how to surface the most useful recommendations. For banks, ML can leverage similar math to identify the most useful data points and weight them properly while ignoring the irrelevant stuff. With proper monitoring of the operation of these algorithms, lenders can begin to identify important shifts in consumer and macroeconomic data over time. That can help lenders avoid sudden turns in the economy.
Conventional credit scores get more expensive. In addition to annual ‘normal’ price increases, some of the biggest players in credit scores are rolling out larger ‘special’ hikes opportunistically when contracts end. One company's executives told Wall Street analysts on public calls that they have already raised prices for mortgage underwriters and got partway through their auto clients in 2019.
That’s going to hit lenders in 2020. Expect more underwriters to begin questioning the soundness of relying on largely generic credit scores compared to leveraging customer-specific data points and AI to craft more nuanced credit profiles.
Credit unions embrace ML for better member service. Credit unions thrive on providing their members with great service. ML will help them do that by driving up approvals with no added risk and automating more of the loan approval process, freeing up human underwriters to focus on providing more personalized services. By the end of 2020, according to a Fannie Mae survey of mortgage lenders, 71% of credit unions plan to investigate, test or fully implement AI/ML solutions – up from just 40% in 2018.
For example, ML will allow them to automate their credit and debit card management systems, speeding up the authentication process. An increasing number of small banks and credit unions are even deploying ML in the core business of credit underwriting. Credit unions already have the performance histories and product data they need from their loyal client base to create effective ML models just as quickly or even faster than they do today with traditional modeling. Beyond providing better service to current customers, ML will enable banks to produce stronger predictive matching of the right product or service with the right applicant or target customer.
For financial institutions that want to stay competitive, 2020 is going to be the year ML takes center stage. That Fannie Mae survey showed that in 2018 nearly 40% of lenders weren’t using ML at all, and about the same sized group were just investigating ML. Tellingly, Fannie Mae found that by the end of 2020, only 2% of lenders believed they wouldn’t be using or considering ML at all. By all indications, those projections remain on the mark.