200 AI agents, humans on standby: Inside Ignosis AI’s debt-collection operation

The next time you get a call from your bank or any other financial institution after missing out on a loan or EMI payment, the voice on the other side might be that of an artificial intelligence (AI) agent.

Voice-based AI agents are quickly making their way into the country’s banking, financial services and insurance (BFSI) sector. Numerous startups in the BFSI and adjacent sectors are integrating AI agents into their services.

Ahmedabad-based Ignosis AI, a data intelligence platform, is one of them. While handling customers, the firm’s system can decide when to call, how to follow up and whether to offer a waiver on penalty fees. If the conversation stalls or turns sensitive, a human can join the live call with access to the borrower’s earlier exchanges, co-founder and chief technology officer (CTO) Chintan Sheth said.


“The system decides whom to call, when to call and what to say,” Sheth told ET AI on the sidelines of the Global Fintech Fest.

The company has built its collections operation on around 200 AI agents that use different languages, tones and scripts, Sheth added. Of the collections lifecycle of four to five lakh borrowers a month Ignosis manages, AI handles about 70-80% of the cases without a human intervention, he said.

Also read: Voice AI startup Arrowhead eyes fresh funding to accelerate international expansion

Founded in 2022 by Nirav Prajapati and Sheth, Ignosis began with infrastructure and analytics for the Account Aggregator framework, which allows people to share financial information with their consent. In September 2025, the company raised $4 million in a round led by Peak XV Partners’ Surge programme, with participation from Force Ventures, Razorpay Ventures and Cred founder Kunal Shah.

Over the past 18 months, Ignosis has expanded into AI products for collections and credit-risk underwriting for small businesses. Sheth said roughly 20 to 25 large accounts are live across the two operations.

Deciding when to call

Lenders give cases to Ignosis just as they would to a collection agency, sometimes before a payment is due, Sheth said. Each case is handled for 30 or 60 days, depending on the arrangement; the firm changes its approach as it learns more about the borrower.

“We take ownership of the outcome during the period we manage the case,” he told ET AI.

The system checks financial records and earlier conversations to understand why a payment was missed. A drop in income, a medical emergency or excessive debt may require a different response from a borrower who routinely pays a few days late, Sheth said.

A strategy agent looks at earlier calls, text messages and WhatsApp chats to decide what to do next. If a borrower promises to pay on a certain date, the system may send gentle reminders. If the borrower has broken promises, it may follow up differently.

Financial data helps the system decide when and how often to contact a borrower, but the AI agent does not receive raw account details during the call.

“We do not expose raw financial information to the large language model or the voice agents. We convert it into internal signals, such as the borrower’s probability of repayment, obligations and usual income cycle,” Sheth explained.

The agent can also negotiate a penalty-fee waiver within preset limits to incentivise the borrower.

The human intervention

Sheth said human intervention happens in 20-25% of the cases, when a human agent joins a call already in progress. The person can see the earlier exchanges and figure out where the AI got stuck.

“If the AI agent cannot progress or the borrower raises something sensitive it cannot handle, a human agent can join the live call within about half a second, with the context of the earlier conversation,” he said.

A death in the borrower’s family or the need to explain the consequences of a missed payment or dealing with a cycle of debt are some of the instances that require such an intervention.

AI’s role in underwriting

Collections is only one of Ignosis’ AI businesses. The company is also using AI to help lenders assess credit risk among micro, small and medium enterprises.

According to Sheth, the system does not approve or reject loans on its own.

“Banks and NBFCs want AI to assist their underwriters, rather than make lending decisions autonomously. Our AI does not approve or reject loans on its own,” Sheth said.

Also read: AI can speed business loans, but underwriters stay in charge: HyperVerge's Kedar Kulkarni

For a small shop, a borrower can share live photos or videos of the shopfront, its surroundings, inventory and cashier. A vision-language model analyses them along with bank statements and GST-based cash flows, and turns it into signals about income, business stability and possible fraud.

The AI flags positive and negative signals for a human underwriter to examine. The underwriter makes the final lending decision, Sheth said.

Many lenders rely mainly on credit-bureau information and some alternative data. This can lead them to reject people with no bureau record or a credit score between 650 and 750 without fully assessing their cash flows and income, he said. He added that bank-statement data could make underwriting decisions three to four times sharper.

Auditing every call

Ignosis sets limits on calling hours, contact frequency and scripts, and tests its agents against various situations before deployment, Sheth said.

“After deployment, a second AI agent audits the call, whether an AI agent or a human made it,” he said.

Lenders retain control through their calling infrastructure and receive recordings, transcripts and outcomes, Sheth said. They can shut off the calling pipeline if they detect an anomaly or use a software connection to stop contact with a particular borrower in real time.

Also read: Agentic payments could test India’s data-localisation rules: Cashfree’s Reeju Datta

As per Sheth, Ignosis takes responsibility for the quality of its agents’ conversations and its role as a collections partner. He also said most clients had seen almost no escalations or complaints to the Reserve Bank of India’s ombudsman.

Across clients, Sheth claimed a 10-40% improvement in recovery within a defined period. Some clients had also reduced their net collection costs by about 25% to 30%, he said.