About eight months ago, a credit head at a mid-size NBFC — two-year FinBox customer, someone I respect — told me: "Rajat, our borrowers are dropping off and we have no idea why. Our RMs are drowning. And every loan file still requires someone to manually chase documents for three days."
Her stack wasn't the problem. LOS, BankConnect, bureau integrations, solid credit models — all in place. The problem was the origination process itself. It was designed for forms, file cabinets, and physical signatures. It had never been rethought for the world her borrowers actually live in — a world where they're on WhatsApp all day and have never willingly filled a 40-field web form.
"The stack is good. The process is broken. We'd been solving the wrong problem."
What the data actually shows
Building infrastructure means you see patterns nobody else sees. Across millions of loan applications on our platform, the drop-off picture is consistent: a significant share of borrowers who start an application never finish it, and the majority of those drop-offs happen at document collection — not because borrowers aren't motivated, but because the interface fails them at exactly the moment it matters most.
The RM productivity picture is equally consistent. RMs spend 80% of their day on document handling instead of selling. They're chasing missing files, manually keying 25–40 documents, and stuck in 2–3 bounce-back cycles per file with credit. That's not a people problem. That's a tooling failure — and it's fixable.
Credit managers have the same problem in reverse. They receive incomplete, unvalidated files and spend their time assembling and reworking instead of doing risk judgment. The result: 21-day average TAT. 30% of applicants dropping off before sanction. Costs that compound every day a loan sits unprocessed.
Why AI changes this — not in the way most people mean
When people say "AI in lending," they usually mean better credit scoring or a chatbot on the website. Neither is the real opportunity.
The real opportunity is using AI to redesign the origination process end to end. An agent that sits at the point of document capture — classifying, extracting, cross-validating, and assembling every loan file autonomously, so it arrives decision-ready at credit. A WhatsApp-native borrower flow that meets people where they already are instead of forcing them through a web form. Intelligence that catches errors and mismatches before they propagate through the system.
This is Atlas. Not AI as a feature bolt-on. AI as the architecture of a better origination process — one where RMs stop chasing and credit managers stop reworking.
What we've actually built
We've been building this for over a year, on top of 10+ years of lending infrastructure. Atlas Origin and Atlas Flow are in active conversations with NBFCs and HFCs. Here's exactly where things stand.
Why we are sharing how we think about this
There's a lot of AI skepticism in lending right now — and most of it is earned. The gap between what vendors claim and what survives contact with a real NBFC's origination flow is significant. Compliance gets handwaved. The "AI" turns out to be a rules engine. The demo looks nothing like production.
We're not interested in that game. Withfinbox.ai is where we share the specifics — the architecture decisions, the testing philosophy, the things that didn't work first time. Not to perform transparency, but because the institutions that understand what's genuinely hard about AI in lending are the ones worth building with.
If you're running or building a lending operation and you want to engage with what we're building rather than just be sold to — the door is open.
