October 29, 2026

By: Intellect

Indian banks ended September 2025 with a gross NPA ratio of 2.2%, a fresh multi-decadal low. In the same report, the RBI notes that the agriculture sector’s ratio, though improving marginally, remains much higher than other sectors, and that agriculture’s share in aggregate bank credit contracted over the year.

Read as a CXO, those facts describe a quiet retreat. When one book carries more credit cost than the rest, and a lender can’t reliably tell which borrowers will repay, the rational move is to lend a little less at the margin. The margin is exactly where the mandate points: small and marginal farmers hold approximately 76% of agricultural credit accounts. The RBI also observes that NBFCs have minimal exposure to agriculture, which leaves the risk largely with banks.

The result is a book that is large, policy-priced and hard to read. This piece argues that the deciding variable in that book is no longer funding or distribution. It is how well a lender can see the borrower.

The score is a shortcut, and the shortcut is straining

A bureau score summarises borrowing history. A large share of farm borrowers has little or none, so the score ends up filtering out the unscored along with the risky. Those are different groups, and a lender that can’t tell them apart loses the first group to protect itself from the second.

Policy has been telling us this for three years. Maharashtra directed banks in June 2024 to do away with CIBIL scores when granting crop loans, repeated the instruction in May 2025 while noting that banks continued to ask for scores despite repeated instructions, and said again in May 2026 that crop loan sanction should not be linked to a farmer’s CIBIL score.

A directive that has to be repeated each year says something that goes beyond one state. Banks reach for the score because it is the one consistent, cheap signal they have, and nobody has given them a replacement they trust enough to defend to an auditor. The CXO’s question is therefore not whether the score should be used. It is what replaces it, and who signs off on that.

The collateral is leaving, so the judgement has to carry the loan

Under the RBI’s KCC Directions, 2026, banks waive collateral and margin for agricultural loans up to ₹2 lakh per borrower, and marginal farmers are eligible for a flexible limit of ₹10,000 to ₹50,000, assessed by the bank without relating it to the value of the land.

The direction of travel is clear. The basis of the loan is moving from what a farmer owns to what the bank can see about the farm. That shifts risk from the asset to the decision. When the decision is the collateral, the quality of the decision is the quality of the book.

The price compounds the point. Under the interest subvention scheme, short-term KCC loans up to ₹3 lakh are priced at 7%, lenders receive a 1.5% subvention, and prompt repayers get an incentive of up to 3%, bringing the farmer’s rate to 4%. A lender earning a policy-set price has two levers left: cost to serve and credit cost. Cost to serve is largely a process question. Credit cost is an underwriting question.

The limit is a formula, the borrower is not

The same Directions standardise how much a farmer can draw. The limit for a crop season is the Scale of Finance for the crop multiplied by the area cultivated, plus fixed percentages for household needs and farm services, plus insurance premium. Every lender applies the same arithmetic.

What the formula doesn’t settle is how risky a given farmer is. Regulation has standardised the “how much” and left the “who” to the lender. That is where differences between lenders will appear, because the formula is common and the judgement is not.

What there is to see now

Data is no longer the scarce input. More than 10.31 crore Farmer IDs have been created under AgriStack as of 3 August 2026, and the Digital Crop Survey covered over 31.3 crore plots across 648 districts in Rabi 2025-26, with surveyors physically entering each plot to record the crop sown. Lenders on the Unified Lending Interface already draw on digitised land records from eight states, satellite and geospatial data, and dairy-related insights. The RBI even lets the farm-services component of a limit pay for remote sensing and satellite-based crop monitoring.

The scarce input is judgement about what to do with it. A useful way to organise that judgement is in three layers, each answering a different question. Identity and land say who the farmer is and what they are entitled to farm, which guards against fraud and ineligibility. Crop and season say what is growing and how it is doing, which speaks to repayment capacity and to shocks that hit a whole area at once. Behaviour and cash flow, such as cooperative payments, mandi sales and repayment history, say how a person handles money.

The common mistake is letting one layer stand in for the others. A satellite reading tells you about a field. It doesn’t tell you who farms it or what happens to the harvest money. Behavioural data tells you about habits. It doesn’t tell you whether this season will cooperate. The value sits in the combination.

Where the return shows up

Better farm intelligence pays in three places, and they pull in different directions.

The first is growth: approving good borrowers who are screened out today for lack of a score. The second is credit cost: avoiding weak approvals and spotting stress early, since farm shocks tend to arrive by geography and crop, not borrower by borrower. The third is cost of decision: less manual verification. The Directions already allow monitoring of end use through field inspections and/or other means as per the bank’s credit policy, so remote evidence has a regulatory home.

The CXO’s task is to decide which of the three to optimise first. A lender pushing growth will accept more thin-file approvals and needs tighter early warning to go with them. A lender focused on credit cost will tighten selection first and expand later. Treating all three as one objective leads to a model that satisfies none of them.

What can go wrong

The risks are real and most of them are organisational, not technical.

Data is uneven. Even where land records are digitised, legacy errors and disputes remain common, and coverage differs by state. A lender has to treat missing data as unknown, not as bad. Otherwise two equally creditworthy farmers get different outcomes because of where they live.

Consent sits with the lender. The RBI has said that ULI does not store customer data or consent, and responsibility for consent, data use and grievances rests with each lender. An institution that automates a decision needs to be able to show the consent behind the data.

And decisions have to be explainable. The RBI’s FREE-AI report sets out 26 recommendations, including a board-approved AI policy for regulated entities, and the framework applies to institutions using AI for credit assessment, among other uses. A score that nobody can explain to a farmer, a branch manager or a regulator will not survive its first dispute.

How to start

The safest way is sequential. Pick one segment, run the new intelligence in shadow mode alongside the existing decision, and compare the outcomes. Start with limit renewals and enhancements, where the bank already holds behavioural history, before moving to new-to-credit borrowers. Define who can override a recommendation and track how often they do. Then promote the model where it holds.

Four measures tell a CXO whether it is working: approval rates among thin-file borrowers, early delinquency, override rates and time to decision. Together they show whether the model is widening access, holding quality, earning trust and saving time.

The way forward

The lenders who benefit most from the shift away from collateral and bureau scores will be the ones who treat farm intelligence as part of underwriting policy, owned by the credit function and reviewed by the board, not as a pilot run by a data team.

Intellect’s AI-First Agri-Lending Solution treats farmer and farm intelligence as part of underwriting, not an add-on. AI-enabled farmer profiling, qualification and screening sit alongside land, crop and allied activity assessment, so a file reflects who the farmer is, what is grown and how the season is going. Alternative data supports inclusive underwriting for thin-file borrowers, multi-bureau reports are available where history exists, and AI-powered fraud and integrity assessment runs in parallel. Cash flow analysis and offer optimisation turn those inputs into a limit and an offer, Smart Field Investigation brings in field findings where a visit is warranted, and collateral and limit management keeps the file consistent after sanction. After disbursal, AI-powered early warning draws on the same data to monitor the portfolio. Products, policies and workflows are configurable, so the lender sets the decision rules, and 44+ ready integrations covering AgriStack, land registries, KYC providers and credit bureaus bring the data layers in.

To know more about how farmer and farm intelligence can strengthen underwriting at your institution, talk to our experts or book a personalized demo.

Also read: “From Compliance Burden to Competitive Advantage: Rethinking Regulatory Readiness in Agri-Lending”

Barnali Chatterjee

Barnali Chatterjee
SME – Digital Lending
Intellect Design Arena Ltd.

Sources

  1. RBI, Financial Stability Report December 2025, Chapter II: Financial Institutions: Soundness and Resilience
  2. PIB, Kisan Credit Card: Fueling Growth in Agriculture (11 March 2026)
  3. All India Radio, Maharashtra Govt Mandates Removal of CIBIL Score for Crop Loans to Farmers
  4. The Week (PTI), Do Not Ask for CIBIL Scores While Disbursing Farm Loans, Fadnavis Warns Banks
  5. TaxTMI (PTI), Banks Asked Not to Link Crop Loan Sanction to Farmer’s CIBIL Score: CM Fadnavis
  6. RBI, Commercial Banks (Kisan Credit Card Scheme) Directions, 2026
  7. Prime Minister’s Office, Cabinet Approves Continuation of Modified Interest Subvention Scheme for FY 2025-26
  8. BioVoice, India Creates Over 10.31 Crore Farmer IDs Under AgriStack
  9. Daily Pioneer, How Agri Stack Can Catalyse Reforms in Land Revenue Administration
  10. MediaNama, Unified Lending Interface Grows to 64 Lenders, 136 Data Services
  11. RBI, FREE-AI Committee Report: Framework for Responsible and Ethical Enablement of Artificial Intelligence
  12. Legal 500 (Saga Legal), RBI’s FREE-AI Framework: Navigating the Legal and Regulatory Landscape

 

Related Blogs

top

Inactive