Synopsis: In today’s blog, we will understand why the human element continues to be crucial for credit underwriting, despite the process becoming automated over time through AI/ML models.

The advent of artificial intelligence (AI) and machine learning (ML) models in finance has ushered in a revolution in the credit landscape. Many new-age fintechs and neo-banks have automated their credit underwriting and fraud detection processes by utilizing AI/ML-based algorithms to offer better, faster, and bespoke loan products. 

With the consumer demand for “anytime, anywhere” finance intensifying every day, traditional lenders have had no choice other than to play catch-up and transform their legacy infrastructure—at significant costs; some have even chosen to partner up with AI-driven financial companies to provide tailored loans.

So, how have ML-based credit models enabled financial lenders to improve their credit offerings, and will this ring a death knell for the ever-prized human touch? Let’s find out.

Higher Machine Learning Adoption for Superior Credit Underwriting

The days of manual processing of loan applications are almost behind us. As per Nasscom, India is well on its way to adding $500 billion to its GDP on the back of increased AI adoption by 2025, with banking and financial services as one of the key contributors.

With more and more customers coming online and leaving behind their traces all over the digital ecosystem, lenders are harnessing access to new sources of data—data that gets fed into the ML credit models. 

With the Account Aggregator system already online, and Public Credit Registry due for an introduction this year, access to aggregated data will improve even further. This will vastly enhance the credit underwriting process, thus affording underwriters an opportunity to source more business and reduce fraud and collection risk. 

But how do ML-driven credit models achieve that?

1. Making Invisible Credit Metrics, Visible

The conventional lending models have broadly ascertained borrowers’ creditworthiness based on credit scores (CIBIL), an inadequate metric at best. Owing to its overreliance on historical financial data and loan history, many new-to-credit (N2C) individuals and businesses had trouble availing of business loans.

In contrast, ML-based credit models have enhanced the process of credit risk assessment by incorporating other key data metrics, such as current cash flows, social media metrics, property records, retailer purchase data, e-commerce payments, and much more.

As a result, lenders are in a better position to evaluate a borrower’s ability to service and repay the loan—as is being done by many NBFCs. This has substantially improved the access to credit for those for whom CIBIL scores served as a poor reflection of their credit risk.

2. Reining in Defaults

ML-based credit underwriting has also helped reduce the incidences of bad credit, on account of upgraded calculations of default risk. Per a study, ML underwriting models reduce defaults by 75% vis-à-vis traditional models. Another study claimed cost savings ranging between 6-25% over better default risk predictions via ML credit models.

3. Continuous Updation and Refitting

Another advantage of employing machine learning for credit underwriting is how easily these models can be scaled and updated. As new data becomes available, machine learning algorithms adjust the likelihood of default and prevent loan mispricing. Such credit underwriting has a self-learning loop, which reduces the long-term costs of model upkeep and updation. As a result, it reduces the possibility of discriminatory lending as prejudices become apparent.

But, do these benefits portend a diminution in the role of human underwriters? Not per se.

Revolutionizing Credit Underwriting through Human-Machine Collaboration

While to some, it may make ardent sense to completely transition to ML-driven credit underwriting, more gains are made when these models are complemented with human underwriters. As per an Accenture report, banks managed to raise their revenue by 34% between 2018-2022, after investing in human-machine collaboration.

So, how do human-machine partnerships repurpose the role played by human underwriters?

1. Auditing and Monitoring the Credit Models

Sure, machine learning has a self-learning loop, wherein it constantly gets updates as fresh data comes in. However, credit analysts must verify that the data is indeed auditable, all the protocols and criteria are being met, and there is no reinforcement of human biases to ensure the models are working effectively. Additionally, based on the ML model’s performance, some parameters and features may be tinkered with for further fine-tuning.

2. Flagging for Manual Review

Even though AI-based credit underwriting yields great results for standard cases, complex loan products can be a different story. Thus, it becomes imperative for such credit models to flag marginal high-risk cases for further review. Credit analysts armed with years of experience and domain expertise are in a better position to weigh in on the merits of the loan application.

3. Domain Expertise and Interpretation of Models

Human underwriters improve the loan extension process by further interpreting the results generated by ML models. Through their extensive domain-specific knowledge, they are in a better position to identify external factors that might adversely impact a borrower’s creditworthiness, such as the state of the economy or the industry. Hence, they function as a valuable add-on to the machine-driven credit underwriting process.

4. Communication to Stakeholders

Finally, credit underwriting isn’t only about hard facts; it also involves nuanced communication skills. A company needs the effervescent human touch to serve its customers better—something only achievable with human underwriters. Credit analysts can provide a more comprehensive review to the borrower, especially about their creditworthiness and the steps to improve it further. 

To illustrate, hypothesize a situation where a lender rejects a loan application exclusively basing his decision on scores borne by ML credit models. In such a scenario, the Robo-advisor will only provide a dry assessment, mostly related to short-term delinquency that was unmet. However, if a human underwriter is provided with these scores, he will be more idiosyncratic in his dealings.

Human Ingenuity + Machine Standardization: The Optimal Approach

Ideally, an effective underwriting process must automate redundant, repetitive, and time-consuming processes through machine learning while leaving space for human underwriters to focus on taking more strategic decisions. By combining the machine element with a human’s cognitive abilities, credit analysts will be in a better spot to contextualize a machine’s predictions, rather than relying on intuition and implicit pattern recognition.

A human-machine approach can truly transform the loan process by providing a comprehensive assessment of credit risk, thus resulting in better decision-making and outcomes for lenders and borrowers alike.