Essay
From Ledgers to LLMs: The State of Indian Lending
Somewhere in rural Maharashtra right now, a loan officer is sitting across from a group of women borrowers, conducting a personal discussion the same way it was done fifteen years ago. Clipboard. Questions. Intuition. A handwritten note that will get partially transcribed into a system later, if at all.
Two hundred kilometres away, a fintech is disbursing a ₹30,000 personal loan in eleven minutes, entirely on mobile, using bureau data and a behavioral scoring model the borrower never sees.
Both of these are Indian lending in 2026. The gap between them - in process, in data capture, in institutional learning - is the story this piece is about.
India's credit market is large and getting larger. Over $2 trillion in outstanding bank loans. A digital lending segment growing at 30%+ annually. 254 million young people entering the credit economy over the next decade. The fundamentals are not in question. What is less certain is whether the operational infrastructure running this market is ready for what's coming.
I've spent the better part of the last decade at the intersection of credit and data - first at American Express building risk models, and now at LendingIQ working directly with banks and NBFCs across India's lending stack. What strikes me is how much the Indian credit story mirrors the broader India story: structurally sound underneath, moving faster than outsiders give it credit for, but carrying legacy weight that doesn't show up in the headline numbers.
This piece traces that arc - from the directed-credit era of nationalized banks, through the NBFC decade, to what I think lending actually looks like in 2047.
One useful lens before we begin
Lending, when stripped down, is an information problem. Every loan decision depends on how well a lender can gather signals, interpret them, and translate them into a risk position. The structure of Indian lending has changed enormously over the decades, but this underlying mechanism has remained constant.
Informal lenders relied on hyperlocal knowledge. Nationalized banks expanded access but operated with limited signal depth. Credit bureaus introduced standardized histories. India Stack enabled identity and remote onboarding. NBFCs improved speed using alternative data. Each phase improved access or velocity - but the ability to consistently process information at scale remained uneven. That is the gap the current moment is beginning to close.
How we got here
1947–2015 - Building access
The early decades focused on reach. Nationalization expanded the banking footprint into rural India, and priority sector lending ensured capital flowed into agriculture and small industries. The system became broader, though often at the cost of credit quality.
Liberalization introduced competition. Private banks brought process discipline and early forms of risk segmentation. The introduction of credit bureaus, and later India Stack, created the first real foundation for structured, digital lending.
Bank nationalisation, wave one
14 banks nationalized. State controls 84% of branches. Priority sector lending mandated for agriculture and small industry.
Regional Rural Banks established
Expanded geographic reach into districts traditional banks did not serve. Deepened financial access at the cost of further fragmented credit quality.
Liberalisation - the LPG moment
Private banks licensed. HDFC Bank, ICICI Bank, Axis Bank emerge. Competition enters the system for the first time. Interest margins begin compressing.
Retail credit inflects
Home loans, auto loans, credit cards scale. CIBIL launched in 2000. Credit bureaus enable the first systematic risk pricing in India.
India Stack changes the game
Aadhaar, UPI, eKYC. 1.3 billion biometric IDs issued. Digital identity makes remote onboarding viable at scale for the first time.
2010–2019 - Speed becomes the differentiator
NBFCs found their edge in execution - faster approvals, more flexible underwriting, and targeted product design. Segments underserved by traditional banks became accessible. A ₹50,000 business loan in tier-3 Rajasthan. A consumer durable loan in 48 hours for a salaried millennial in Pune.
Even so, much of the decisioning still relied on human interpretation. The system moved faster, but it did not become more consistent or more learnable. Every credit officer carried institutional knowledge that the system itself could not retain.
Where the friction sits today
The scale of what's at stake makes this worth getting right. India's credit market is already enormous - and still early.
None of this growth automatically resolves the operational gap underneath it. A typical underwriting workflow illustrates the problem. A credit officer might spend 30–45 minutes conducting a borrower discussion. Only a fraction of that interaction gets captured in structured form. The rest remains experiential - useful in the moment, but difficult to reuse, audit, or build on.
Across institutions, this accumulates into real operational drag:
- Variability in decisions across similar borrower profiles
- Limited feedback loops for improving models over time
- High dependence on individual expertise that exits with the person
The scale problem compounds this. At ₹50,000 average ticket sizes across millions of borrowers, the economics of deep manual review simply do not hold. Trade-offs get made between speed, depth, and coverage - and quality is what absorbs the slack.
When we work with small finance banks on their microfinance PD processes at LendingIQ, the pattern is consistent. The loan officer carries an enormous amount of signal - their read of group dynamics, their sense of which borrowers are coached, their intuition on household cash flow. That intelligence is not being captured, structured, or learned from at the system level. It retires when they do. The opportunity is to make it persistent.
The emerging shift
What's beginning to change is not just the tooling, but how work itself gets executed across the lending lifecycle.
The concept taking shape is an AI workforce - a set of systems that can take ownership of specific lending workflows, handling data interpretation, applying policy, and producing structured outputs that feed into decisions. These systems operate within existing infrastructure rather than replacing it.
In practice: borrower interactions conducted in local languages while being structured in real time. Loan files reviewed continuously against policy rather than sampled post-disbursement. Collections strategies that adapt based on borrower behavior and past interactions. Early risk signals surfacing before they appear in standard delinquency metrics. The effect is cumulative - decisions become more consistent, processes more observable, and learning more persistent across the organization.
This is the architecture we're building toward at LendingIQ: intelligence layered above the existing stack - the LOS, LMS, bureau APIs - without requiring system replacement. A voice agent conducting a PD in Bhojpuri. An audit agent reviewing disbursement compliance on every file. An ML model flagging portfolio stress before it shows up in the 90+ DPD bucket. Each operates on top of infrastructure lenders already have.
The role of ULI
The RBI's Unified Lending Interface adds an important layer. It enables standardized access to multiple data sources - land records, GST data, insurance histories - through a consent-driven framework. For thin-file borrowers who have historically fallen outside formal credit, ULI meaningfully improves the signal available at underwriting.
The effectiveness of ULI depends on how well lenders integrate this data into actual workflows. Access to data and the ability to act on it are different problems. The institutions that build the operational layer to use ULI well will see the most benefit.
India lending in 2047
In 2047, India will mark 100 years of independence. The Viksit Bharat vision targets a $30–35 trillion economy with per capita income broadly in the upper-middle-income range. The credit implications are significant - a country at that income level has a middle class that borrows for homes, education, and retirement; an MSME sector capable of absorbing structured credit at scale; and a manufacturing base that will need trade finance and supply chain credit products that barely exist in India today.
The history of Indian lending can be read as a gradual reduction in information asymmetry. Each phase improved visibility in some form. The current shift is more internal - less about expanding the system outward and more about how effectively it operates within.
Across three broad horizons, here is how we see it unfolding:
Institutions that integrate workflow-level intelligence begin to see meaningful reductions in cost and improvements in consistency. Execution quality becomes a differentiator. Regulatory frameworks around AI in credit decisioning start to crystallize - DPDP compliance and algorithmic transparency become active concerns, not future ones.
Credit integrates into broader ecosystems - supply chains, enterprise systems, gig platforms. Underwriting shifts toward continuous evaluation rather than one-time assessment. ULI matures into a persistent consent layer, and bureau scores give way to dynamic creditworthiness models updated at every transaction.
Credit allocation becomes a problem of intelligent capital routing. Decisioning frameworks evolve to match borrower risk profiles with investor risk appetites dynamically, across a fragmented but interoperable lending ecosystem. The boundaries between lenders, platforms, and infrastructure layers become genuinely porous.
What remains constant
Some characteristics of Indian lending are likely to persist regardless of pace. Trust continues to be built locally - particularly in semi-urban and rural geographies, ground-level lending remains intensely relational. Regulation plays a central and often non-linear role; the RBI has consistently shown it will intervene when credit expansion outruns risk management capability. And the 300+ million thin-file borrowers in India today will not all be fully served by 2047 - but ULI and alternative data should meaningfully shrink that number for lenders who build the underwriting capability to serve them profitably.
Closing
The history of Indian lending can be read as a gradual reduction in information asymmetry. Each phase - nationalization, liberalization, credit bureaus, India Stack - improved visibility in some form.
The current shift is more internal. Less about expanding the system outward and more about how effectively it processes information within. The institutions building this operational layer - structured data capture, workflow intelligence, continuous learning - are the ones likely to define what Indian lending looks like at its centenary.
The lenders who will be relevant in 2047 are building that infrastructure now.