Lack of access to affordable bank credit for MSMEs has remained a perennial issue for not just the entrepreneurs but also for lenders and the government alike to bring businesses in the informal economy into the formal fold. Poor credit access emanates from the lack of collateral with MSMEs to pledge against the credit. To tide over this challenge, over the past few years, banks and non-banking financial companies (NBFCs) have been gradually integrating historical data of MSMEs including GST filings, cash flow information, and more to underwrite business or working capital loans to them.
The ease of access to this data coupled with the use of artificial intelligence (AI) powered algorithms to analyse the repayment ability of the borrower has enhanced lenders’ dependence on such rule-based credit underwriting models. This translates into quicker disbursals and faster growth of loan portfolios for lenders.
However, the Reserve Bank of India (RBI) now wants such lenders particularly NBFCs, whose reliance on rule-based credit engines is relatively higher than banks, “to recognize that rule-based credit engines are only as effective as the data and criteria upon which they are built. Overreliance on historical data or algorithms may lead to oversights or inaccuracies in credit assessment, particularly in dynamic or evolving market conditions,” said Swaminathan J, Deputy Governor, RBI on May 15 at the Conference of Heads of Assurance of NBFCs.
Cautioning NBFCs over lending based on historical data, Swaminathan said there appears to be a fancy among most NBFCs to do more of the same thing, such as retail unsecured lending, top-up loans or capital market funding.
“Over-reliance on such products may bring grief at some point in time later. It is also observed that the risk limits that are fixed for certain categories of products or segments, say like unsecured lending, in some entities, is way too high to be sustainable in the long run.”
NBFCs, cautious on their part, agreed to ensure a balance between AI and human intelligence-led models instead of over-dependence on the former.
George Alexander Muthoot, Managing Director at Muthoot Finance said the RBI has rightly advised against setting up high risk-limits for unsecured loans.
“A mere scorecard generated by AI-based credit models may not capture the nuances of the MSME landscape or accurately predict future performance,” Muthoot told FE Aspire.
In fact, AI-based credit models, said Muthoot, can be one component of a broader assessment strategy which also includes traditional evaluation methods such as personal discussion (PD), field visits, and document verification
While objective parameters in underwriting a loan can be automated to a great extent, evaluating subjective parameters will require a human element. This is because although these statistical and machine learning-based data models provide strong insights, their assumptions are based on historical data, which may not hold true once market conditions change.
Hence, MSME lending models today, said Nishith Maheshwari of InCred Finance, are not just based on historical data, such as the bureau which largely only has a record of previous repayment behaviour.
Maheshwari, Head, Digital Business Loans, InCred Finance told FE Aspire that with visibility of GST filings, transactional/cash flow data, customer data based on online feedback and reviews for a product, etc., models can build a lot more insight into both the financial and performance metrics of businesses.
Moreover, experts believe that NBFCs lending to MSMEs while cautious of potential risks are also confident of having checks in place to avoid any error and grief in future.
Tirthankar Datta, Partner at law firm JSA Advocates and Solicitors told FE Aspire NBFCs lending to MSMEs have to inevitably find more advanced algorithms which would be able to dynamically adjust itself based on the data which can be constantly collected by it.
This may mean that NBFCs potentially will have to find a niche use case for an intuitive user interface or experience on mobile applications, which are being preferred to traditional banking channels by younger customers. Here, virtual reality and augmented reality may be one area which is ripe for exploitation for finding new niches, according to Datta.
So what new models can be developed as an alternative to existing models and also be error-free?
Vikrant Narang, Deputy CEO of Ambit Finvest told FE Aspire that a good start would be to develop a model by identifying parameters based on expert judgment. Statistical tools and AI/ML-based models could be used to test the evidence of the weight of these parameters.
Moreover, NBFCs may find ways to access data on the unbanked MSMEs or lower-income individual citizens by gaining access to non-traditional sources of data, which may not be available in the mainstream credit data repositories. According to Datta, this could be in the form of mobile payments, payment gateway and UPI payments-related data which could be a gold mine for enriching existing data sets.
Importantly, RBI governor Shaktikanta Das earlier this year at an event had also highlighted the likely risks of depending on algorithm- or model-based lending, FE had reported. The governor had cautioned lenders on “the ground realities that keep on changing. So, it is important to monitor whether your model is falling behind the curve or is in tune with the times, and what are the possible risks.”