By Abhishek Gupta, Co-Founder, Dekrypt Labs
For decades, credit analysis at Indian NBFCs followed the same playbook. A credit analyst would pull financial statements, run ratios manually in Excel, call a few contacts for ground-level intel, check CIBIL scores, and write a Credit Appraisal Memorandum (CAM) that would sit in an approval queue for days. The process was thorough — but painfully slow and deeply human-dependent.
That's changing fast.
In the past 18 months, a growing number of Indian NBFCs — from mid-sized retail lenders to microfinance institutions — have started deploying AI-powered credit intelligence tools to automate parts of this process. The results are hard to ignore: faster turnarounds, more consistent risk assessments, and the ability to underwrite borrowers that traditional models would have flagged as "thin file."
Here's what's actually driving this shift.
A typical credit analyst at an Indian NBFC spends 6–12 hours on a single CAM for an SME loan. They're pulling MCA filings, cross-referencing GST returns, checking for litigation exposure, validating promoter backgrounds, and summarising industry risk. Most of this is grunt work — structured data extraction that doesn't actually require human judgment.
The judgment part — the "should we lend?" question — gets about 20% of the analyst's time. The other 80% is data collection and formatting.
This isn't just inefficient. It introduces inconsistency. Two analysts looking at the same borrower often reach different conclusions based on how much data they managed to pull, what they chose to emphasise, and frankly, how tired they were when they wrote the report. For a portfolio of thousands of loans, these inconsistencies accumulate into real credit risk.
Modern credit intelligence platforms don't replace the credit analyst — they eliminate the 80% of work that shouldn't require a human in the first place.
Here's what a well-designed system can do in minutes that would take an analyst hours:
Company intelligence aggregation. Pull and parse MCA filings, GST registration status, CIBIL commercial data, court records, and news mentions — automatically, for every borrower in the pipeline.
Financial statement normalisation. Convert unstructured CA-certified financials into standardised ratio analysis, flagging anomalies like sudden revenue spikes or working capital compression.
Promoter background checks. Identify directorial overlaps with defaulter companies, related-party exposure, and beneficial ownership structures that are deliberately obscured.
Industry benchmarking. Compare the borrower's ratios against sectoral medians — something most analysts skip because the data is hard to access quickly.
Regulatory signal monitoring. Flag if the borrower's sector has recently attracted RBI or SEBI scrutiny, or if there are pending policy changes that could affect repayment capacity.
All of this feeds into a draft CAM that an analyst reviews, amends, and signs off on. The human doesn't disappear — they just stop doing spreadsheet work.
India's lending market has unique characteristics that make AI-powered underwriting especially valuable.
First, data fragmentation. Borrower information in India is scattered across MCA21, GST portals, court databases, income tax records, and sector-specific regulators. No single API gives you everything. Manual aggregation is the only option — unless you build or buy a system that does it for you.
Second, the SME credit gap. India has roughly 6.3 crore MSMEs, and formal credit penetration remains below 15%. A major reason lenders avoid small businesses is the cost and risk of underwriting them manually. AI-driven CAM automation brings that cost down by 60–70%, making SME lending economically viable at scale.
Third, regulatory pressure. RBI has been tightening its expectations around credit risk frameworks. Boards and audit committees now want documented, consistent, auditable underwriting processes. AI-generated CAMs with clear data sources and logic trails are actually better suited to regulatory scrutiny than analyst-written ones, where the underlying data is often undocumented.
It's worth being direct about limitations. AI credit intelligence is only as good as the underlying data. India's public databases — particularly MCA and court records — are incomplete, inconsistently formatted, and often several months stale. Any system claiming 100% accuracy on borrower intelligence is lying.
What a good AI system does is surface data gaps explicitly, flag confidence levels, and ensure that analysts know exactly where the information came from. That's a different problem than the black-box models that got a lot of NBFCs into trouble with algorithmic lending in the 2018–2020 period.
The goal is augmented intelligence — faster, more consistent, and more auditable — not autonomous lending decisions.
The NBFCs that are moving fastest on AI underwriting aren't the largest ones — they're the mid-tier lenders who have enough scale to feel the pain of manual processes but enough agility to adopt new tools without a two-year IT procurement cycle.
Over the next 24 months, expect AI-powered credit intelligence to become table stakes for any NBFC operating in the SME or consumer lending space. The cost advantages are too large to ignore, and RBI's push for more robust credit risk frameworks is only going to accelerate adoption.
The credit analysts who thrive in this environment won't be the ones who resist the tools — they'll be the ones who learn to work with them, focusing their energy on the judgment calls that AI genuinely can't make.
If you're building or operating a lending business in India and want to see how AI-powered credit intelligence works in practice, visit dekryptlabs.com.