Kogta Financial is a retail-focused NBFC providing vehicle finance, loans against property, and secured business loans to MSMEs across 300+ branches for nearly three decades. Kogta has served more than 100,000 customers, including farmers, transporters, taxi operators, first-time borrowers, and small business owners.

Debt collection is one of the hardest problems in the banking sector. It is labor-heavy and spread across dozens of languages and dialects. Borrowers span distinct languages and communication norms, and a one-size-fits-all script wasn't cutting through.

Kogta wanted to move faster and adapt its voice collection workflows in real time. With the majority of its technology already built internally, the company decided to build and own its voice collections stack end to end, giving its teams greater control and faster turnaround times.

Choosing Smallest AI

Kogta wanted to build the voice stack in-house, where it had the most control: build its own agents, edit the call flows itself in real time, and ship changes in hours to start seeing results the same day. After evaluating the options, Kogta chose Smallest AI for a few reasons.

The quality: the agents sounded natural enough that borrowers couldn't tell the difference, and accurate enough to trust with real accounts.

The pace of execution and support: when Kogta flagged an issue or asked for a new feature during the initial setup, it shipped in hours. And Smallest worked the KPI alongside Kogta's own team rather than from a distance.

"With Smallest, it never felt like a vendor and a client. It was always equal ownership of the outcome, and that was the best part."
- Vishal Handa, CTO, Kogta Financial

Building a voice stack Kogta owns end to end

What stood out first was the speed. Kogta moved from POC to full production in two weeks, starting with early-stage collection buckets. Within days, it expanded to harder buckets and began collecting payments across them.

The integration was entirely API-based. Kogta could track agent performance directly through the platform’s analytics. When the team needed a change or hit a blocker, Smallest AI often turned it around within hours.

Under that sits one decision that mattered more than any single feature: the entire stack runs on Smallest AI's own voice models. Pulse STT improved accuracy across rural accents, mixed-language speech, and numbers spoken in regional languages. Lightning v3.1  handled text-to-speech, giving the agent a natural, human-sounding voice on every call. 

For some agents, switching from GPT to Electron SLM V2 cut latency by nearly half, making conversations faster and reducing awkward pauses between turns.

"Smallest owning the entire stack changed everything for us. Pulse gave us the accuracy we needed on rural accents and mixed languages. Electron SLM trained on our languages made the conversations far faster and more natural. This is the real game-changer for us."
— Chandan Agarwal, CPO, Kogta Financial

The impact: real money, on the call

Kogta has multiple call recordings where the agent pushed the conversation and got the borrower to pay on the spot, including big-ticket EMIs worth lakhs of rupees, not just the low-value ones people assume voice agents are limited to.

• 13M+ collection calls placed

• ~50% lower latency after moving to Electron SLM

• 130+ concurrent calls across multiple agents at peak, at a 0.2% failure rate

Borrowers didn't treat the agent like a bot. Because they knew every detail about the borrower's loan information and could answer whatever they asked on the spot, most came away believing they had spoken to a real Kogta representative.

The agent was able to access each borrower’s loan details and answer questions in real time. Many borrowers believed they were speaking with a Kogta representative, not a bot.

The effect carries down the whole collections chain. It doesn't stop at the call: the system reads each borrower's intent, and by end of day the insights roll up and flow down to Kogta's field team, so they know exactly which accounts to act on next, and how.

The reach is widening too. Kogta has begun deploying multilingual agents across more regional languages, and can see the results in real time as each one goes live.

Compliance held up at scale. In lending, this is the part that matters most. Collections is one of the most tightly regulated conversations a lender has, and getting it wrong carries real regulatory risk, so a black box was never an option. Smallest is built to be a compliance-ready platform for regulated industries, and conversations stay inside RBI collection norms.

• Full visibility and control over what the agent says on every call

• The whole regulatory structure managed in-house

• Conversations kept inside RBI collection norms

• Every call logged and auditable, start to finish
  
  

"In collections, compliance is the whole game, and Smallest is compliance-first by design. Every call runs on a stack we control end to end, so we can see exactly what the agent says and keep it inside RBI collection norms. That's what made us comfortable putting real accounts on Smallest AI agents."
— Abhishek Chordia, CISO, Kogta Financial

What's next

Collections agents were the proof point for Kogta. Its ambition now is to deploy agents across the whole customer journey. Wherever Kogta talks to a customer, it should eventually happen through a voice agent.

"I feel voice AI will be the first touchpoint for every customer interaction at Kogta. We start with the hard use cases and then move to the other functions of the business."
— Vishal Handa, CTO, Kogta Financial

That expansion is already underway, with agents now handling calls in Kogta's loan origination system, and the early results are coming in.

As the system evolves, Kogta will keep expanding what the agents can do, enabling them to handle more actions, support more use cases, and reach more customers through fast, natural voice conversations, until voice is the first touchpoint for every customer interaction.