Tavant Vice President Atul Varshneya outlines the role of artificial intelligence and machine learning in improving the credit decisioning process in consumer lending and why the sector needs to move faster with the adoption of such innovations.
Let us start with the definition of the term artificial intelligence (AI). Minsky and McCarthy, the fathers of this field, described artificial intelligence as any task performed by a program or a machine that, if a human carried out the same activity, we would say the human had to apply intelligence to accomplish the task. While this is a fairly broad definition, it conveys the idea in a simple and approachable way. It is, therefore, not surprising that this is indeed the way the term AI is spoken of and understood in most industry circles.
However, to be a bit more specific, what AI is used to refer to in almost all practical cases in the industry, is actually machine learning (ML). Machine learning is a narrower concept compared to the broader idea of AI as defined above. It is specifically about using data, most of the time a lot of data, and to extract latent information from it without being explicitly programmed. We as humans see patterns in our experiences and are able to discern commonality among them and are then able to infer various aspects through those. For instance, based on a person’s facial expression we can infer if she is happy, sad, angry, etc. Machine Learning techniques enable similar capabilities in machines, such as identify patterns in the data to detect probability of fraud, propensity to accept a promotion offer, likelihood of a borrower defaulting in paying a mortgage loan, etc.
ML in consumer lending
It is evident that within the mortgage industry the interest in machine learning is currently rising significantly as most players are recognising its power in helping with efficiency, speed-up, as well as accuracy, in various business processes, many of which are at the core of their business. For instance, ML models can bring up in the decisioning process specific parts of the loan application that require closer scrutiny, as opposed to the entire application; this can make the process faster and less error prone. Similarly, models to predict churn can help with improving retention.
The diagram below shows examples of such use cases across the life of a loan. While in general, the industry has jumped in a big way into the idea of machine learning and its potential value in their business, we see that most companies in the mortgage ecosystem are still experimenting with ML and have not rolled it out widely. The largest players who build most of their technology in house are probably the furthest along.
Some processes are indeed data-driven but use simpler linear statistical techniques, such as investors’ credit risk models. Using machine learning methods here can significantly improve the accuracy of predictions of loan and mortgage insurance performance (defaults, prepayments, etc.). The availability of large amounts of data to train the ML models is gradually becoming less and less of a stumbling block. These methods can be used both to price riskier loans more appropriately as well as provide loans to borrowers who would otherwise have been turned away by overly conservative risk models.
Other business processes have typically been driven by hand-written rules, such as portfolio retention. Here, the use of ML models can make an even bigger difference by learning from actual outcomes automatically (e.g., who actually paid off their loan and who accepted a retention offer) and do so with greater precision than human intuition.
Finally, some business processes today are often not data-driven at all in many companies, such as the process of converting leads to loans. Here the use of ML can have a big impact in improving conversion ratios and reducing the costs of working on loans that don’t close.
Using machine learning methods can significantly improve the accuracy of predictions of loan and mortgage insurance performance
In all of these areas, the application of ML has benefits beyond just improving decision accuracy. ML also inherently adapts to the changing business environment through continual learning as new data flows in. Furthermore, augmenting human decision-making through the use of ML brings higher consistency in such decisions and also reduces the time to make those decisions.
Wide adoption of machine learning in the consumer lending industry is inevitable. The question is not if, but when, it will cross the threshold of being indispensable. The industry needs to spend time and effort on identifying when it will occur and begin planning and executing actions to prepare. Players that are early with a strategy to prepare for it will be the leaders of the transition from today’s work environment to that of the future, and will benefit from the competitive advantage it brings. Players without this planned transformation are likely to find themselves trying to play catch-up or worse.