Traditional sources of data are no longer sufficient for a comprehensive risk analysis of MSME and retail customers. Instead, alternate data, including cash flows, social media footprint, transaction analysis, industry analysis and proximity to economic activity hotspots can enhance a customer’s credit worthiness. AI is helping such solution providers get there faster. Abhishek Singh, Chief Analytics Officer at Lendingkart explains how.

How does LendingKart leverage AI?

Lendingkart objective has been to enable the financial institutions to support young and educated digitally savvy entrepreneurs who are based in non-metro cities and have growing micro enterprises, across a wide range of industries. The traditional data sourcing methodologies for conducting analytics have been mostly focused on underwriting using an expert scorecard or score cards based on logistic regression. Traditional scorecards are based on bureau data and demographics superimposed with policy rules. While explain ability is very high in traditional scorecards, techniques like logistic regression are not able to capture non-linear relationships in data and hence a large segment of credit worthy customers are missed out from the capital access. Alternate data, including cash flows, social media footprint, transactions analysis, industry analysis, and proximity to economic activity hotspots, can help assess customers’ creditworthiness.

Lendingkart uses AI-ML tools to analyse the bank statements in a very advanced way to get trends around credit, debits, balances, cheque returns. Access to non-traditional data sources such as SMS data, location information, platform interaction information along with bureau variables form an important part of the information, and their non-linear correlation allow us to identify profiles of customers that are potentially high performing but otherwise would have been missed in traditional methods. These stacks have evolved from traditional logistic regression to models such as random forest or gradient boosting methods one is able to capture the non-linear relationship between the predictive variables

How does AI actually help your customer?

Our algorithms allow evaluation of New to Credit (NTC) customers. A layer of interpretability has been created using shapely values, to ensure ML algorithms are no longer the black boxes they used to be. Advanced AI-ML techniques have strengthened our risk assessment capabilities along with developing hyper customized products for our customers. We have identified various proxies for determining the customer’s capability, intent, and habit to pay back a loan. Additionally, AI tools are leveraged across the funnels to evaluate the quality of his product/service, financial health of his business, and the ability to survive with competition to take a holistic view of the customer.

Could you tell me more about your explainable AI, its features and how it helps the business & consumer?

For a business, especially in new segment, explainability of its AI model decision becomes important for all stakeholders to operate efficiently at every stage. As explained earlier, our AI algorithms use many features that interact in non-linear ways, and it becomes even more essential to drill down precise factors that are impacting the risk profile of a business. For instance, a customer defaulting on payment instruments categorises him high risk, whose credit eligibility will be impacted. Business teams are continuously seeking reasons of a particular borrower being profiled under a particular risk category and their credit eligibility. AI model will give the output in terms of eligibility and risk profile. Given Lendingkart’s history as a pioneer in cash-flow based lending and consistent high growth, the need for explainability of model predictions is very high. Xlr8 is the origination platform where explainable AI has contributed in a big way to prioritize, pre-screening, express assessment to optimize the resources being spent at the top of the funnel. Our customers come to know about their eligibility digitally within seconds and don’t have to visit branch or submit paperwork. This helps them to plan their credit requirement and opt for credit product that is best suited for their business. The explainable AI also capacitates our origination channel partners to optimize their efforts towards the best fit customer segment. Cred8 is the Credit Intelligence platform that has been built for partner financial partners to leverage Lendingkart AI underwriting prowess, which is available for partners to use once they are onboarded.

Today’s MSME borrower is getting data rich as data collection avenues have expanded. As an NBFC, how does data collection and verification happen, and how does AI aid this process?

One of the important things that have happened with the advent of technology is that one is able to track the customer journey from end to end. With the digital India campaign, development of India Stack Aadhar-based customer authentication, Digital Signatures, Digital Payments via UPI, and adoption of digital technologies has not just boosted business opportunities for Indians, but also transformed the way they operate. Introduction of Goods & Services Tax (GST) further helped simplify business taxes and increase tax reporting. The GST acts have achieved a milestone in formalizing the MSME community with GST registration of approximately 9.2 million businesses all over the country. The online registrations and tax reporting has created a large pool of digital data from the MSMEs – additionally, data that is verified, updated and accessible electronically, thus reducing the frauds and related risks. The potential game changer has been ‘India Stack’ - which has accelerated this digital evolution of Indian economy. India Stack with other additional APIs has become a rich reserve of both public & private data.

To give a boost to this new age lending platform capabilities, RBI released the concept of “NBFC Account Aggregators” addressing the need to have an open data sharing infrastructure in place where the data could be extracted from the financial information provider directly based on explicit customer consent, via a safe and secure platform. It can not only ensure instant access to data required for underwriting but also eliminate data submission frauds.

ML techniques combined with technology to collect data is now enabling companies on multiple fronts - 1. Approve customers that would not have been otherwise approved based on additional data 2. Assess the risk and price customers appropriately 3. Optimize marketing and operations cost. Use of modern data mining techniques has facilitated in designing better products for the customers and reduced the cost of various financial products as operating expenses will reduce considerably with easy and instantaneous access to data and digital delivery mode.

How can data mining for fintech be refined for future use cases and business offerings? What other use cases are you working on right now?

The model development has been a comprehensive journey from manual interventions to closing the feedback loops to automated underwriting algorithms. The model ensemble has transitioned in past 5 years and the major benefits can be seen at various stages:

• The approvals rates trends
• having a dynamic risk-based pricing and exposure strategy
• Early Warning Indicators as well as portfolio benchmarks and any anomaly, industry changes, trends are fed back into the model learnings
• New Normal: The Covid19 pandemic and lockdowns throughout the country have resulted in delayed ITR filings and lowered efficacy of the credit bureau data (6 months of moratoriums resulting in stagnation of bureau information, further compounded by One Time Restructuring Offers)
• Customer prioritization and appropriate strategy at origination, delivery and servicing stage based on the data trends that are fed into AI model

The use cases that are being worked right now for future are on Collection Intelligence platform that offers digital connect, communication, field coverage and predictive modeling of the EMI dates as per the receivables and payables of the business. The off-book performance and alternate data will augment the core variables in credit decision making along with eco system capabilities such as geo location tagging, social data etc. to further refine the product features being offered, such as enhancing the credit line facility and making it dynamic based on performance and business predictions. This involves expanding the alternate data and further analyzing phone apps, frequency of usage can be an indicator of the risk profiling of the customer and these data capture the ongoing economic trends faster than direct information from the customer.

Did COVID-19 impact the efficacy of your algorithms, given that there could have been delays in loan payments among borrowers?

The Covid pandemic sparked much apprehension in the outlook of the economy, as the initial projections of spike in the economy took a major setback due to complete focus on saving lives and livelihoods amidst a once-in-a-century crisis. After two quarters of lockdown, the situation had finally started to improve and the MSME sector was bullish about getting back on their feet in the new normal. Lendingkart model algorithms have consistently been updated for simulation of these factors and multiple scenarios in its self-learning underwriting model which has strengthened our risk assessment capabilities and build robust early warning systems for such testing periods of time. Hyper customized products have been developed and driving hyper customized products for our customers in partnerships with corporates by integrating APIs to offer best suited financial products to MSMEs like supply chain financing, marketplace lending, and cross-sell products in coming times to best recoup post this wave subsides.

What does the state of fintech today tell you about the average MSME borrower?

More than 70% of Indian population resides in rural areas, which have been increasingly becoming part of ‘Global Village’, because of advent of technology through smart phones and internet connectivity. The era of villages being remote has passed and virtual connectivity has been placing the rural population at similar footing in terms of business as city suburbs when it comes to visibility, communications, sourcing, and competitive product pricing.

MSMEs are generally stereotyped as hesitant or slow to adopt technology, preferring traditional and conventional style of business dealings and its operations. However, in the last few years, more small business entrepreneurs have been adopting digital technology in their operations boosting their efficiency and reach out multi-fold. It has initiated a circle of improved technology and other services for them, and hence more businessmen are moving from traditional ways to digital methodologies with prevalent best practices available at their disposal through wide information available on the internet.

The new generation of entrepreneurs are being encouraged and facilitated with the flexibility and simplicity of executing business that comes with digital technology. These entrepreneurs can start their business online and deliver their products worldwide or even provide services to a global audience. High data connectivity at affordable prices and increased penetration of electronics (especially smartphones, computers, tablets) has further accelerated the digital adoption process for MSMEs. MSME sector is one of the most poised sectors ready for next age innovation as it will be one of the major steppingstones to make India a 5 trillion-dollar economy in coming years. With the ecosystem developments such as Account Aggregator, the Indian MSME story is in making for the world to witness and get inspired by.