Kogta Financial, one of India's fastest-growing NBFCs, was losing collections battles because of rigid scripts, language mismatches, and vendor lock-in. With Smallest AI's voice agents, they went from signed contract to full production deployment in five weeks automating 96.2% of collection calls, cutting latency by 50%, and collecting EMIs worth hundreds of thousands of rupees, directly on-call.

When borrowers won't pay

Collections at scale is a people problem before it's a technology problem. For an NBFC managing billions in vehicle and MSME loans across India's semi-urban belt, getting borrowers to respond and pay on a call requires more than volume it requires trust. And trust starts with language, familiarity, and the feeling that the person on the other end actually knows your account.

Kogta Financial had all the scale. What they lacked was the agility to match it. Their borrowers span Rajasthan, Gujarat, Haryana, and Maharashtra each with distinct dialects, communication norms, and expectations. A one-size-fits-all Hindi script wasn't cutting through. Engagement was low. Collection rates were stuck. And every time the team needed to tweak an agent script or test a new approach, they had to call their vendor. No self-serve. No speed. Pure dependency.

Their KPI was never soft metrics no "promises to pay," no "follow-up scheduled." The measure of success was simple and unambiguous: real money collected on the call.

Built for real ownership

Smallest AI builds voice AI infrastructure for enterprises that need production-grade agents not demos. The platform is designed for teams who want to own their agents, iterate in real time, and integrate machine-to-machine without manual overhead.

For collections specifically, the stack handles the full conversation: live EMI lookups, dynamic responses to borrower questions, dialect-aware delivery, and real-time escalation logic all without a human in the loop.

When Kogta came to Smallest AI, the pitch was simple: build your own agent in days, go live in weeks, and measure success the same way your call center does by money collected.

Road to live collections with Voice AI in 5 weeks

Kogta signed in December. Within five weeks, the first live EMI-bounce calls were running in full production on the Smallest AI platform. No CSV uploads. No manual handoffs. Pure machine-to-machine integration, triggered by Kogta's own ML systems deciding which accounts to call and when.

Vishal Handa, CTO of Kogta, built the first agent himself in three days.
What followed was a rapid iteration cycle. Issues surfaced after the first EMI cycle were fixed before the next one. The Smallest AI engineering team including the founders were on WhatsApp, not in an email thread.

Agents launched first on the DPD 0–30 bucket early-stage collections where speed and tone matter most. Then came the harder buckets: DPD 30–90, where the stakes are higher and the conversations are tougher. Agents handled both. EMIs worth hundreds of thousands of rupees were collected directly on-call not just small-ticket reminders.

As scale grew, Kogta moved their agents from OpenAI's LLM to Smallest AI's proprietary Electron SLM. The result: a 50% drop in response latency, with no degradation in conversation quality. In fact, the conversations got sharper responses came so fast the team had to tune for overlap, not lag.

96% automated. Every day.

Five weeks after signing, Kogta had a fully automated, multi-agent collections stack in production.

• 96.2% of collection calls now handled entirely by voice AI

• 300,000+ minutes of collections automated per month across the fleet

• 50% lower latency with Electron SLM vs. OpenAI — no drop in borrower experience

• Agents deployed across DPD 0–30 and DPD 30–90 buckets simultaneously

• New use cases already live: telephone verification (TVR), loan origination calls, and lead qualification

What comes next for Kogta

The roadmap goes further. Kogta's vision is that every customer conversation from onboarding to collections to qualification eventually runs through voice AI. The collections stack was the proof point. The rest of the customer lifecycle is next.

"Every duplicate subscription, every auto-renewal nobody caught: it adds up before you see it coming. It’s not a failure of any one person on the team. It’s a failure of the process."

Vishal Handa,
CTO, Kogta Financial

This is what production-grade voice AI looks like: not a pilot that lives in a spreadsheet, but a system that runs at scale, collects real money, and gets better every cycle.