The Reserve Bank of India’s move towards an Expected Credit Loss (ECL) framework marks a significant shift in how banks assess credit risk, moving from a reactive approach to one that anticipates potential losses across their loan portfolios. The transition will require stronger data capabilities, sharper risk models and continuous monitoring to identify signs of borrower stress before they translate into defaults. In a conversation with Deccan Chronicle, Rajesh Kumar, Chief Executive Officer of Lentra, discusses what the new framework means for banks, the role of AI and predictive analytics in identifying emerging risks, and how technology can help lenders make more informed decisions across the loan lifecycle.
Rajesh Kumar, Chief Executive Officer of Lentra.
Excertps
What is the ECL framework, and how will it change the way banks assess credit risk and provide for potential loan losses?
When we talk about Expected Credit Loss (ECL), the fundamental difference is that we are looking at risk across the entire loan book, including the good book.
There are three components: Probability of Default (PD), which measures the likelihood of a borrower defaulting; Exposure at Default (EAD), which measures how much would be outstanding; and Loss Given Default (LGD), which estimates the loss after recoveries.
Banks also need to continuously monitor every loan and classify it under a three-stage provisioning model.
First and foremost, ECL is a proactive approach. With sharp models that accurately measure risk, banks can define, manage and calibrate their risk appetite more effectively. By comparing predicted defaults with actual outcomes, they can course-correct their estimates, credit policies and pricing.
Which banks will come under the framework, and how will implementation challenges differ depending on their loan portfolios, data quality and technology readiness?
This is a landmark reform in risk assessment. Under the master directions released by the RBI, the ECL framework applies to all Scheduled Commercial Banks, public and private.
The implementation differences will largely come down to data quality and analytical capabilities. PD, LGD and EAD models require accurate, relevant data, supported by application and behavioural scorecards that capture changes in borrower risk.
Banks will differ in how well they capture data and connect information across systems to build a 360-degree customer view. They also need timely information on relevant macro and micro changes.
Having said that, data itself is not enough. The analytical skills to build, deploy and dynamically fine-tune models as economic conditions and credit cycles change will drive successful implementation.
A borrower may continue paying EMIs on time even as their financial position weakens. What signals should banks track to identify this deterioration before repayments are affected?
A borrower paying EMIs on time does not necessarily mean their financial position is healthy. If cash flows are weakening, the borrower may be taking on new leverage to repay existing loans. That can lead to loan stacking and overleveraging.
We need to watch early warning signals around two things: cash flows and leverage. The point is to spot stress before it shows up in repayment behaviour, giving banks a chance to act proactively.
Once a bank identifies an early warning signal, how should that insight translate into action, from borrower engagement to changes in underwriting and portfolio strategy?
Any monitoring framework is only as good as the alerts it raises and what the bank does with them. The real discipline lies in ensuring last-mile action, followed by feedback loop to constantly improve effectiveness.
A short-term cash shortfall, say a large customer paying late, may call for closer monitoring and a conversation with the borrower. But similar signals across an industry may require revisiting credit policy, exposure limits and the bank’s appetite for that segment. Communication strategies may also need to change as repayment behaviour evolves.
Where can AI and predictive analytics meaningfully improve credit-loss assessment, and how should banks validate these models and explain their decisions?
Credit stress builds through several small shifts across different portfolios, which can be missed when each team looks only at its own data. An institution-wide intelligence layer can identify these patterns across lakhs of accounts and give the bank enough insights to effect quick interventions.
This requires continuous back-testing against actual outcomes and calibrating models across customer segments and economic conditions. Explainability matters just as much. Every decision should be traceable, fair and supported by a clear record of the information and approval process behind it.
Where does Lentra fit into a bank’s ECL preparation? Can you illustrate how its lending technology and credit analytics could help turn an emerging risk signal into a lending or portfolio decision?
For almost a decade, Lentra has pioneered enterprise platforms for lending. Today, we work with over 30 lenders across secured and unsecured lending, and our platforms have processed over 40 million loan applications.
We bring expertise across credit, risk, technology and analytics, supported by more than 500 data signals and a platform hosting 60 AI models. The acquisition of Data Harbor has further strengthened our capabilities in building credit-risk models.
We are also building AI platforms such as Viveka to connect data across systems and bring institutional knowledge into decision-making. Together, these capabilities position Lentra to support timely, proactive decisions across the loan lifecycle.
Disclaimer : This story is auto aggregated by a computer programme and has not been created or edited by DOWNTHENEWS. Publisher: deccanchronicle.com
