Introducing Artificial Intelligence (AI) into the mortgage industry is by no means a new concept. AI technology has been around since the early 1990s and became widely accepted by 2005. While advancement slowed during the great recession as mortgage companies sought new ways to de-risk, the last two years have seen AI explode in the financial sector.
No longer the terrifying replacement to human labor that Stephen Hawking long predicted, industry professionals hope AI will increase margins, decrease the burden on originators, and limit repurchase risk. AI has also begun to benefit the consumers by improving user experience. In the mortgage industry, it can take a high-stress, unfriendly process and make it almost frictionless.
Recently, the first major partnership in finance over AI was brokered less than a year ago between IBM and Fannie Mae who hope to use the technology, known as Watson, to lower default rates and prevent future crises.
But how exactly does AI accomplish all this? Lenders work with an AI “expert system” that requires 3 things. (1) A set of rules, (2) a set of data that is a question solved by the rules, and (3) a user interface which will filter the data to conform to the rules in the system. In the current system underwriting standards are the rules, the data comes from client applications, and the underwriting is the user interface.
Instead of needing a highly specialized underwriter, AI is able to receive infinite amounts of data and translate that into a risk profile. This is done by running tremendous amounts of this data through an algorithmic model until it finds enough patterns to be able to make decisions on the data. Using prediction analysis AI supposedly can determine the optimal rate to charge a particular client.
This model is already in use for credit scores. Firms look at thousands of variables including social media, interest history, geolocation, and other smartphone data to determine the optimal credit score. In terms of mortgages, AI is a first line of defense against risky borrowers. When an AI system flags a loan, it is sent to a loan officer for final review.
When trained correctly, AI is significantly better than humans at detecting fraud as well as previously undetectable defects in the lending system. The problem AI often encounters is that the data that it needs to properly determine default risk is complex, unorganized, and “noisy.”
This data dilemma is the main reason why AI is still not widely used in the mortgage industry. Currently we are unable to produce the quality of data needed to develop and properly train AI systems. Lenders are wary of model accuracy after AI models backed many of the loans granted during the great recession.
For now it appears that the current lending model is just starting to embrace a new era of AI technology. However, little efforts are being made to change management and human functions as well as the proper oversight to manage data. In the coming years AI will be used as a first line defense against fraudulent financial transactions and as a way to quickly screen borrowers. As AI begins to influence more and more of the finance sector, mortgage origination will eventually be forced to enter the digital era.