By Varun Malhotra, Chief Technology and Data Officer, Capri Global Capital Limited.
Artificial intelligence is revolutionising the lending industry where the work is document-heavy and data-driven. It can support lenders in checking information, identifying risk signals and assisting borrowers more efficiently. The strongest model, however, is not autonomous lending. It is a meticulously regulated blend of AI, fixed rules, and responsible human judgement.
Optical character recognition can retrieve data from tax records, bank statements, invoices, and identity documents, while language models can structure unstructured text and identify inconsistencies. Tasks that once needed hours of manual reading can now be finished in minutes, enabling credit teams to concentrate on analysing data.
AI is reshaping the lending workflow
Optical character recognition can extract information from bank statements, tax records, invoices and identity documents, while language models can organise unstructured text and highlight inconsistencies. Tasks that required hours of manual reading can be completed in minutes, allowing credit teams to focus on interpreting data.
AI can also make the first level of assessment more consistent. Two reviewers may interpret the same bureau record differently because of experience, workload or context. A governed model can surface relevant data points in the same sequence for every applicant, reducing subjective variation. Historical data, weak proxies or incomplete information can reproduce unfair outcomes. Consistency therefore has to be paired with bias testing, data-quality controls and a route for borrowers to seek review.
For businesses, AI can combine cash-flow patterns, seasonality, sector trends, local conditions and repayment behaviour. It can flag falling collections, unusual invoices or mismatches between declared turnover and account flows. The model should organise evidence; the credit officer should decide how much weight patterns deserve.
Digital rails can turn analysis into a better borrower experience
India’s connected digital ecosystem allows analysis to flow into origination, verification, servicing and disbursement. With the customer’s consent and safeguards, lenders can combine digital identity, document repositories, account information, e-signatures and payment systems into a paper-light journey. Applicants can receive clearer status updates and timely requests for missing information instead of repeated branch visits and opaque hand-offs.
AI can help personalise interactions, summarise terms and direct a borrower to the right support. Customer information must be checked against approved product terms, and sensitive data collected only for a defined purpose. A seamless journey must remain accurate, secure and understandable.
Rules still dominate where policy must be deterministic
Lending will continue to rely on rules-based automation. Regulatory and policy checks require clear outcomes: whether mandatory documents are present, exposure remains within limits, a product is permitted for a borrower segment, or an application requires enhanced due diligence. These controls need predictable behaviour, a policy owner and auditable reasons. Rules enforce known boundaries. AI interprets information, recognises patterns and ranks signals. A sound architecture uses rules as guardrails, AI to improve assessment, and human review wherever the cost of error is high.
Human oversight matters most at the edge cases
Borrowers do not always fit historical patterns. A small business may show seasonal revenue declines. A self-employed applicant may have uneven income but strong repayment discipline. A young enterprise may have limited history yet credible contracts and improving cash flows. The question is whether a recommendation is appropriate for the applicant, not whether a model is accurate.
High-risk, high-value and unusual cases should trigger meaningful human review. Credit officers need underlying evidence, explanations of influencing factors and authority to challenge recommendations. Overrides should be reasoned, recorded and analysed. Justified overrides may reveal missing model features; unexplained overrides may indicate control problems.
Borrowers should receive clear reasons when credit is declined or offered on different terms. Alternative data can widen access only if it is relevant, proportionate and open to challenge. This is particularly important for small businesses, first-time borrowers and people with limited formal credit histories, who need a route to seek review.
Accountability must be designed into the system
The 2026 Global AI in Financial Services Report from the Cambridge Centre for Alternative Finance found that 81% of surveyed financial services firms were adopting AI at some level, and 40% reported advanced adoption. Yet the same study identified data privacy, unreliable outputs, model opacity and loss of human oversight among the leading risks. The gap between adoption and governance is now a central management issue for lenders.
Responsible deployment requires ownership of models and data, validation, monitoring for drift and discrimination, escalation thresholds, and records allowing decisions to be reconstructed. Accountability must extend across technology, business, credit and risk teams. Not every process benefits from a probabilistic model, and no model should make irreversible decisions merely because automation is possible. Success should reflect not just speed and cost but a responsible, fair and transparent assessment. Governance must advance with innovation. Technology can support better decisions, but accountability must remain with people.
Reference:
2026 Global AI in Financial Services Report – Adoption, Impact and Risks – CCAF publications – Cambridge Judge Business School. Adoption and risk figures cited in the article.