How ML-Driven Early Warning Systems Help Lenders Monitor MSMEs The technology itself has evolved over the past decade to cater to the processing of structured and unstructured data at scale by leveraging ML-driven technology

The Reserve Bank of India (RBI) has released Master Directions for compliance by public and private banks and select FIs for monitoring of early warning signals. RBI and the department of financial services (DFS) have defined a comprehensive framework listing 42 and 83 signals, respectively, to be generated based on the data obtained from these sources.

An increasing thrust from the government bodies to open up regulatory, tax and other relevant data repositories to the public has made the implementation of EWS a reality. Also, the technology itself has evolved over the past decade to cater to the processing of structured and unstructured data at scale by leveraging ML-driven technology.

A new-age EWS solution powered by ML

It has now become imperative for lenders to have an automated solution to track borrower's health in a near-real-time given the deteriorating credit quality and the overall uncertainty in the economy in the wake of the current COVID-19 crisis. An EWS solution that is powered by ML is on alert 24x7 and can pick up events indicating leading signs of distress on borrowers. It integrates with a multitude of data sources, from both the public domain and within the bank. The system then ingests this data and performs AI-/ ML-based analytics to issue the early warning signals as and when they occur so as to enable the lender to classify the account as Red Flagged Account (RFA).

An ML-powered EWS is scalable and can continuously monitor the portfolios and share alerts/notifications on a daily basis. The notifications are directly sent to the customer via email and further the triggers can get captured as action items in the case management module where tickets are assigned for further resolution.

Why ML-driven early warning signals are the way forward

Over 60 million micro, small and medium enterprises (MSMEs) in India have sparse data availability. One of the major challenges for lenders is the time and effort required in monitoring the MSME portfolios for risks. Monitoring only traditional financial data points is not relevant and it requires analysing multiple disparate sources of alternate data as well to form a 360 view on borrowers. The cost of monitoring such long-tail MSME portfolio customers is high. An automated EWS eliminates the challenges of manually handling MSME portfolios with a large number of borrowers.

• Ability to manage huge volumes of portfolios
• The AI-/ML-driven prediction models can pick up leading signals of distress 12-18 months in advance of the actual precipitation of risk. The models pick up on faint signals from sources such as GST an EPFO
• NLP models process unstructured data such as news, social media, and more such alternated data sources for sentiment analysis
• An ML-powered EWS suppresses false-positives by learning over time and setting thresholds at optimal levels

Reducing the false-positive rate with continuous learning

One of the main properties of ML is that it can keep learning if data is fed into it. This means the system improves as time passes, sets thresholds at optimum levels, and issues more accurate alerts.

Over time, as algorithms improve, the false positive rate will continue to decline, driving significant cost reductions to operations. ML and AI can quickly flag changes in patterns of activity, whether caused by new product offerings or money laundering.

Review unstructured data by training NLP models

Banks are often sitting on piles of data which is unstructured and hence is not useful to the credit decision-making process. The amount of this kind of unstructured content is accelerating at an unprecedented rate, making it time-consuming to analyze. As a result, unstructured content is underused as a source of insight. The overwhelming volume of data makes it impossible to spot the nuances that could drive a decision-making process. Machine learning can not only cleanse but also restructure this data to retrieve important information. Natural language processing (NLP) offers opportunities to uncover meaningful insights from under-used content. Based on this information, early warning signals are generated and lenders can identify potential loan defaulters early on.

No scope for human bias

When done manually, the monitoring process involves the risk of human bias. Over the past few years, we have come across incidents where human bias has caused financial losses for banks. With EWS powered by ML, the scope for any human bias can be eliminated, thereby facilitating completely accurate credit decisions.

The Indian banking sector has long been facing the strain of bad debts. With the pandemic further adding to their woes, it is only natural that they are apprehensive of lending to MSMEs. With an ML-powered EWS in place, however, banks can protect themselves from the clutches of bad debts while being able to extend their credit services.