With rising early-stage defaults in micro, small and medium enterprises (MSMEs) loans, particularly from private banks, industry experts have said that traditional credit models, largely based on past financial statements, collateral value and static credit scores, are failing to capture early MSME risk.

Industry insiders warn that conventional methods are not capturing early distress signals like missed bridge financing or slight working-capital shortfalls, resulting in a blind spot for emerging risk. In response, several leading private lenders are tightening eligibility criteria and reducing sanction limits by up to 15 to 20 per cent, aiming to curb exposure until more forward-looking risk models are developed.

“Rising early-stage defaults in MSME loans, particularly among private banks signal that traditional credit models may no longer be sufficient in assessing risk. These models typically rely on historical financials and collateral, which often fail to reflect the dynamic, seasonal, and volatile nature of MSME operations. As India shifts towards becoming a digital economy, there’s a growing need to move beyond static assessments and adopt real-time, data-driven models,” Kumar Shekhar, Deputy Country Manager, Tide in India.

A joint report by TransUnion Cibil and the Small Industries Development Bank of India (Sidbi) revealed a 13 per cent year-on-year (YoY) growth in total MSME credit exposure, reaching Rs 35.2 lakh crore as of March 2025. However, it also pointed to a 14 per cent decline in fresh loan originations by private banks and a rise in early-stage defaults among low-ticket loans.

“The keyword here is ‘traditional’. These credit models you are referring to were built for credit assessment of a significantly different kind from what banks and financial institutions need now, especially when it comes to MSMEs. So, it is little wonder that data suggests these models are unable to capture early risk warnings. ‘What’ has to be assessed has changed, and therefore, the ‘how’ has to as well,” said Raja Debnath, Chairman, Co-founder and Chief Executive Officer (CEO), Veefin.

In Q4 FY 2024–25, the portfolio-level delinquency rate for MSME loans fell to 1.79 per cent, the lowest in five years—yet this masks troubling early-stage defaults that are not yet classified as non-performing. Despite the improving headline ratio, private banks have reported a notable pullback in new MSME disbursements.

“There’s growing evidence that traditional credit models are falling short, especially in identifying early stress among MSMEs. They rely heavily on past repayment data and bureau scores, which are often missing or delayed. AI fills this gap by analysing real-time signals like GST filings, current account flows, and business invoices. This enables earlier detection of risk and sharper differentiation between viable and vulnerable borrowers,” stated Joydip Gupta, APAC Head, Scienaptic AI.

Can AI Bridge The Gap?

While traditional underwriting relies heavily on retrospective financial data, AI-led credit assessment models are showing early promise in improving MSME risk detection. These tools also helped reduce turnaround time for loan decisions. However, experts caution that AI adoption remains uneven across the sector and outcomes depend heavily on data quality, explainability, and regulatory oversight.

“Many of these issues are being successfully addressed by AI. Alternative data is a term used in almost every discussion about the evolution of credit underwriting – and with good reason, too. AI-led underwriting offers lenders access to, and insights from, several sources of alternative data – bank statements, GST data, etc. – to make an informed decision on a borrower’s creditworthiness. The traditional models you referred to need an AI shot in the arm to help them learn both what has to be evaluated and how. That’ll go a long way in bridging the MSME credit gap,” Debnath stated.

Notably, AI, while not a silver bullet, is increasingly being viewed as a complementary layer to strengthen MSME credit infrastructure rather than a wholesale replacement of traditional methods. There is a risk of algorithmic bias is real, especially when models are trained only on bureau-centric data. Additionally, many lenders still hesitate to trust AI outputs fully.

“There’s still some hesitation because of three main reasons: limited explainability, lack of regulatory guidance, and internal change resistance. For risk teams to trust AI, they need clarity on how decisions are made, assurance that the models are auditable, and proof that outcomes improve portfolio performance. As more Indian lenders see reductions in early defaults and better inclusion, that trust is growing,” Gupta told BW Businessworld.

Shekhar added that his is a temporary phase. As AI systems become more transparent and regulators offer clearer guardrails, we believe trust will grow significantly over the next two to three years. New-age lenders, who are digital-native, are already showcasing the power of AI in expanding credit access responsibly. Their success will serve as a catalyst for traditional institutions to accelerate adoption, balancing innovation with accountability.

Human Touch Remains Crucial

Thin-file MSMEs, often informal or newly registered, pose a unique challenge. With little traditional data available, AI models draw from alternative sources such as bank statements, GST filings, digital payments, e-invoicing records and behavioural data. However, experts warn that over-reliance on AI without human oversight risks alienating creditworthy borrowers who fall outside the model’s parameters.

AI is powerful at pattern recognition and risk stratification, but in thin-file or high-ticket MSME loans, human judgment remains essential. The ideal approach is to use AI to handle the heavy lifting — initial assessments, fraud checks, early warnings - and reserve human review for edge cases. This hybrid model combines efficiency with empathy, especially in markets where a loan denial can mean business closure, Gupta added.

Notably, AI models used by lenders are also yet to fully integrate the operational complexities of key policy schemes like the Credit Guarantee Fund Trust for Micro and Small Enterprises (CGTMSE) and the Emergency Credit Line Guarantee Scheme (ECLGS), both of which aim to de-risk MSME lending. Experts noted that AI should serve as a tool to augment human judgement, not replace it, especially in high-stakes MSME financing, where a denied loan can often mean a business shutting down.

"The current AI infrastructure in MSME lending – much like in most other areas – is evolving. It can be prone to not fully capture the significant eligibility criteria and complexities of schemes like CGTMSE and ECLGS. While AI excels at processing vast data quickly, integrating detailed policy requirements such as collateral guarantees or sector-specific conditions remains challenging. Many AI models focus primarily on creditworthiness metrics and may miss subtle program rules without customised training and updates," Debnath stated.

Bridging The Data Gap Needs Unified Action

“Collaboration among tech providers, credit bureaus, regulators, and lenders is essential. This includes building robust, bias-aware data architectures, establishing shared ethical frameworks, ensuring transparency through explainable AI models, and enabling continuous monitoring. However, current efforts fall short due to siloed development, lack of standardised practices, limited public engagement, and prioritising functionality over safety. Additionally, data privacy and security challenges remain insufficiently addressed. Strengthening cross-sector cooperation and standardisation can help in accessing AI’s full potential for fair and inclusive credit assessment,” Debnath added.

While AI-based credit decisioning has made inroads into MSME lending, its success ultimately depends on the strength of the ecosystem supporting it. Experts argue that without collaboration across institutions and standardised data practices, the promise of fair, inclusive and scalable credit delivery will remain unfulfilled.

“We need stronger collaboration across the ecosystem tech platforms must integrate with public infrastructure like Account Aggregator and GST; credit bureaus need to improve access to SME data; and regulators should create clear frameworks for AI use in credit. Today, efforts are too fragmented. For AI to fulfil its promise, the ecosystem must move from parallel innovation to coordinated action,” Gupta mentioned.

Experts point out that although policy and infrastructure frameworks such as the Account Aggregator, GST network, and digital public infrastructure are in place, they are not yet fully integrated into AI systems used by lenders. More seamless coordination is needed for technology, data, and regulatory pathways to converge and truly enable inclusive lending at scale.

“Lenders, in turn, must commit to integrating these technologies responsibly. Today, the gap lies in fragmented data flows, inconsistent interpretation of compliance norms, and limited interoperability between systems. A more aligned ecosystem with shared standards, open APIs, and real-time policy integration can accelerate trust, reduce biases, and expand credit access meaningfully, especially for underserved MSMEs,” Shekhar stated.

Without ecosystem-wide synchronisation, AI risks amplifying existing inefficiencies rather than solving them. As India looks to scale up credit access for its 6.3 crore-strong MSME sector, the push must now shift from experimentation to integration.