Tejas AI Wants to Kill the Guesswork in Bank Credit Decisions

The Macro: Credit Decisioning Is Stuck in 2010

Here’s something that surprised me when I first learned it. Most banks, even large ones, still test credit rules by running them against historical data in spreadsheets or clunky internal tools, then waiting weeks or months for model validation teams to sign off. The process for deciding who gets approved for a loan, what interest rate they pay, and what credit limit they receive is shockingly manual at many institutions.

This matters because bad credit rules cost banks real money in two directions. Too loose, and you’re approving borrowers who default. Too tight, and you’re rejecting profitable customers who go to a competitor. The sweet spot is narrow, and finding it requires the kind of rapid experimentation that most bank risk teams simply cannot do with their current tooling.

The vendors in this space are a mix of legacy players and newer entrants. FICO has dominated credit scoring for decades, but their decisioning platform is expensive and slow to implement. Experian PowerCurve, Provenir, and Zest AI all offer pieces of the credit decisioning stack. GDS Link and Scienaptic AI are going after the same mid-market bank segment. But the common complaint from risk analysts I’ve talked to is the same across all of them: the feedback loop between writing a rule, testing it, and deploying it is too slow.

What AI specifically brings to this problem is the ability to simulate outcomes at scale. Instead of testing a rule change against last quarter’s data and hoping it generalizes, you can model thousands of scenarios, stress-test edge cases, and get a much better read on how a policy change will actually perform before it goes live.

The Micro: Simulate Before You Ship

Tejas AI, a Y Combinator W25 company, is building what they call a risk decisioning platform for banks. The founder, Gaurav Luhariwala, previously led product and business at Tartan, where he scaled revenue 8x in a year building KYC and data solutions for Indian banks. That’s relevant context because it means he’s seen the inside of bank procurement and compliance processes, which is where most fintech startups die.

The core product pitch is about eliminating uncertainty in credit rule development. If you’re a risk analyst at a bank and you want to change the debt-to-income threshold for auto loan approvals, today that’s a multi-week process involving data pulls, backtesting, committee reviews, and maybe a champion/challenger test in production. Tejas wants to compress that into something closer to a simulation environment where you can model the impact of rule changes before committing to them.

Think of it like a sandbox for credit policy. You define your rules, the platform runs them against your data, and you see projected outcomes: approval rates, expected default rates, revenue impact, regulatory compliance flags. The value isn’t just speed, it’s confidence. Risk teams make better decisions when they can see the downstream effects of those decisions before they hit real customers.

The platform appears to sit between the data layer (where customer information and credit bureau data lives) and the decisioning engine (where rules get executed in production). That’s a smart positioning because it means Tejas doesn’t have to replace the bank’s existing core systems. It’s an augmentation layer, not a rip-and-replace.

I find the timing interesting. Banks are under pressure from both regulators (who want explainable models) and fintechs (who are eating their lunch on approval speed). A tool that lets traditional banks iterate on credit policy faster, without sacrificing the auditability that regulators demand, addresses both pressures at once.

The question I keep coming back to is integration complexity. Banks have notoriously fragmented data infrastructure. Getting clean, unified customer data into any analytics platform is often the hardest part of the project, and it’s a problem that Tejas can’t solve for them. Their success is going to depend heavily on how well they handle the messy reality of bank data.

The Verdict

I think the problem is real and the market is big enough. Credit decisioning is one of those banking functions that everyone agrees needs to be faster and better, but that most vendors make harder than it needs to be.

At 30 days: who are the first customers? If Tejas is landing community banks and credit unions in the U.S. or mid-size banks in India, those are very different go-to-market motions with very different integration requirements.

At 60 days: how deep does the simulation go? The difference between “we run your rules against historical data” and “we model macroeconomic scenarios and predict how your portfolio performs under stress” is the difference between a nice-to-have and a must-have.

At 90 days: what does the regulatory picture look like? Bank regulators are still figuring out how they feel about AI-assisted credit decisioning. If Tejas can position itself as a tool that makes decisions more transparent and auditable, that’s a feature. If regulators see it as a black box, that’s a blocker.

The credit risk space is notoriously hard to break into because banks move slowly and trust is earned over years. But the founders seem to understand that world, and the product is aimed squarely at a pain point that risk teams feel every day. I’m cautiously optimistic.