Pin-code flood, heat and cyclone scores across an Indian lender's loan book
TL;DR Perilbook scores every address in a lender's loan book for flood, extreme heat and cyclone exposure, then rolls the scores up into the tables a risk committee and the RBI's climate disclosure draft ask for. Mid-size banks and NBFCs pay a yearly licence. It works because global tools price Indian locations coarsely.
Indian lenders know bad weather hurts repayment, but they rarely know which branches, pin codes or borrowers carry the most exposure. Germanwatch's latest Climate Risk Index places India among the countries hit by repeated extreme events over 1995–2024[1].
Location matters a lot more than averages suggest. A 2026 study of 424 Indian power plants found unprotected flood losses averaging above 19% of asset value, with tail losses near 58% in Kerala. A 25-year flood defence cut the average to about 3.3%[2].
So two loans in the same district can carry very different risk depending on elevation, drainage and protection, and a district-level heat map simply cannot show that. Most risk teams have nothing finer.
Perilbook takes a lender's loan tape (addresses or pin codes, outstanding, sector, tenor) and returns a hazard score per loan for flood, heat stress and cyclone wind. Scores combine public rainfall, elevation and coastline data with past event records, and they're checked against local damage reports. Results land in a dashboard and an export that maps straight onto disclosure tables - share of the book exposed, by hazard and by severity.
The seed pitched a general hyper-local climate model. I reframed it, because a model with no buyer attached is a research project. Lenders have a dated reason to pay, so the product starts there and moves to insurers and housing finance firms later.
What it does, in regulatory terms - risk analytics only. Perilbook does not rate borrowers' credit or make lending decisions, so it needs no RBI licence. Lenders stay responsible for their own disclosures and credit models.
The chief risk officer's team uses it and the same lender pays. Scheduled commercial banks, top and upper layer NBFCs and all-India financial institutions are in scope of the RBI draft[3]. Global vendors don't publish prices[4][5]. My assumption is ₹10–30 L a year per mid-size lender, priced by loan count, with a cheaper one-off portfolio scan as the way in.
The RBI's February 2024 draft asks covered lenders to report governance, strategy and risk management from FY2025-26, and metrics, including assets vulnerable to climate risk, from FY2027-28[3]. Do confirm the final text on the RBI site. Either way, risk teams have two to three years to build numbers they can't produce today.
Nobody publishes a figure for Indian spending on physical climate risk data. The narrowest signal is the RBI draft's list of covered entities - every scheduled commercial bank, Tier-IV urban co-operative banks, five all-India financial institutions and top and upper layer NBFCs[3]. Each one needs a defensible exposure number. Next door, one US climate-analytics firm says three of the five largest US banks buy its data[4].
US climate-analytics firms sell asset-level, multi-peril projections to banks and insurers[4]. One Australian firm screens portfolios and supports central-bank stress tests, though its site doesn't mention India[5]. Big consultancies run one-off climate studies. In my limited search I found no Indian specialist that scores loan books at pin-code level with local damage data. The gap is local accuracy at a price a mid-size NBFC can pay.
| Difficulty | Moderate (44/100) |
| Starting capital | ₹60–120 L |
| Time to launch | 9–12 months |
The modelling is doable with public data. The hard part is proving accuracy against real damage records, getting a risk team to share a loan tape and turning scores into numbers auditors will accept. Expect each sale to take two or three quarters, and banks being banks, maybe longer.
No licence is needed for risk analytics, but your buyers are regulated and will treat you as a vendor.
Building this outside India? The categories above usually have a local equivalent, so check with your own regulator.
AI shouldn't feed credit decisions on individual borrowers here, and the scores are only as good as the hazard data behind them.
Want to work out the AI layer for Perilbook? Talk to Yash →
Here's what can go south -
I think this is worth a shot for a founder who has worked in credit risk or catastrophe modelling and can sit across from a CRO. A pure data-science team will struggle to sell it. First, prove that your scores explain past defaults or restructurings after one real flood better than district averages do.
Plan for roughly ₹60–120 L, mostly data scientists, hazard data and long sales cycles with banks and NBFCs.
No, as long as you don't rate borrowers' creditworthiness. Expect vendor audits and DPDP duties as the lender's processor.
Yes, and global firms already sell it. Local hazard data and disclosure rules are what make it useful in each market.
The free Perilbook validation kit has the line-by-line capital breakdown, the effort scoring, the rules and data checklist, a 14-day validation sprint with pass/fail targets, a 90-day launch plan, 10 interview questions for risk teams, and 5 assumptions to test. Download the kit →
Thinking about building Perilbook or something close to it? The make-or-break work is showing one lender that your scores track its own losses better than what it uses today. That's the part I like helping founders with. If you need a hand, you know where to find me. Talk to Yash →
For illustration only. Names are fictional and ideas hypothetical; sourced facts are real but may be out of date. Nothing here is legal, financial or professional advice, and Yash Khatri is not personally liable for any use of it. Read the full disclaimer.