Ahmedabad-based Ignosis AI says it is using roughly 200 voice-based AI agents to manage collections for 400,000 to 500,000 borrowers each month—handling 70% to 80% of cases without human intervention. When a conversation becomes sensitive, disputed or unusually complex, the company routes the live interaction to a human agent who enters with the full conversational history already in view.

For the collections industry, the model offers a concrete picture of what “agentic AI” may look like in practice: not simply a chatbot or predictive dialer enhancement, but an operating layer that prioritizes accounts, selects a contact strategy, conducts multilingual borrower conversations, monitors performance and escalates exceptions to people.

A Collections Operation Built Around AI

Ignosis describes its platform as an end-to-end collections operation for early-bucket accounts. Lenders can submit cases through an API or spreadsheet, after which the company’s AI tools develop borrower-specific engagement strategies using prior interactions, repayment behavior and borrower-consented financial-data signals available through India’s Account Aggregator framework.

According to Chintan Sheth, Ignosis co-founder and chief technology officer, the system determines “whom to call, when to call and what to say.” It can alter the cadence and tone of outreach, decide how to follow up on a promise to pay and, within lender-defined limits, offer a waiver of penalty fees as a repayment incentive.

The company says its approximately 200 AI agents use different languages, tones and scripts. Its website says the collection platform supports more than 12 languages and more than 50 personas designed for different borrower circumstances and delinquency scenarios.

The architecture appears designed to move beyond mass outbound calling. Rather than treating every overdue borrower alike, the system attempts to classify contactability, payment intent, affordability and risk, then identify the next best action. For example:

• A borrower who generally pays several days late may receive a reminder timed to an expected income cycle.

• A borrower who has made and broken several promises to pay may receive a different follow-up strategy.

• A borrower with financial stress, a dispute or a request for a settlement may be routed to a human specialist.

• A borrower with available liquidity may be directed toward a digital payment option instead of a higher-cost field or agent intervention.economictimes.indiatimes+1

Ignosis says it can also use Account Aggregator-derived signals to time mandate presentations around likely cash-flow and balance windows, with an asserted ability to predict debit success in the next 24 to 72 hours.

Humans Remain the Exception Handlers

Although Ignosis presents the platform as highly automated, the human handoff is central to its operating design. Sheth said human intervention occurs in about 20% to 25% of accounts, often when the AI cannot advance the conversation or the borrower raises an issue the system should not handle alone.

That may include a death in the family, a medical emergency, questions about the consequences of delinquency or situations involving a borrower’s broader debt cycle. In those instances, Ignosis says a human can join the active call in roughly half a second and see the prior conversation, rather than receiving a cold transfer without context.

This distinction matters for agencies, lenders and compliance teams evaluating AI. The operational objective is not necessarily to remove humans from every interaction. It is to reserve human capacity for cases where judgment, empathy, negotiation skill, documentation or regulatory sensitivity matters most.

Ignosis’ own materials frame the division of labor similarly: AI agents handle high-volume, lower-complexity follow-ups, while people are assigned disputes, nuanced settlement discussions and high-risk conversations.

Data Signals, Not Raw Account Data

A key feature of the company’s approach is its claimed use of financial-data intelligence to guide outreach. Ignosis began as an infrastructure and analytics provider for India’s Account Aggregator system, which enables consumers to share financial information with consent. Founded in 2022 by Sheth and Nirav Prajapati, the company later expanded into AI-driven collections and MSME credit-risk underwriting.

Sheth said that while the platform uses financial records and interaction history to infer factors such as repayment probability, obligations and typical income timing, it does not provide raw account data directly to the large language model or voice agent during the call. Instead, the system converts that information into internal decisioning signals.

That separation is significant. Collections models that incorporate granular bank or cash-flow information face heightened questions about data minimization, borrower consent, model governance and whether an AI interaction could inadvertently disclose personal financial details. Turning underlying data into constrained signals may reduce some exposure, but lenders and collection partners would still need to establish clear policies around data permissions, permitted uses, retention and auditability.

Ignosis says it provides calling-hour limits, contact-frequency controls and defined scripts, and that it tests agents against various scenarios before deployment. The company also says a second AI agent reviews each call—whether handled by a machine or a human—for quality and compliance.

Compliance Is the Real Test

The scale and reported economics are attention-getting. Across clients, Sheth told The Economic Times that Ignosis has seen recovery improvements of 10% to 40% during a defined management period, along with net collections-cost reductions of about 25% to 30% for some clients. Those are company-reported performance claims, not independently verified benchmarks.

The company’s platform materials cite additional claimed results, including a 4% to 5% improvement in overall efficiency, 25% more collected through digital channels, a 10% improvement in contactability and a 20% reduction in escalations.

For U.S. creditors and agencies, the more important takeaway may be the governance framework rather than the headline productivity figures. Any comparable deployment would need controls that map to the Fair Debt Collection Practices Act, Regulation F, Telephone Consumer Protection Act, state debt-collection laws, privacy obligations and creditor-specific policies. Core controls should include:

• Clear limits on call frequency, calling windows and channel use.

• Reliable detection and suppression of cease-communication, attorney-representation, dispute, bankruptcy and deceased-consumer indicators.

• Strong controls over payment arrangements, settlements and any fee-waiver authority.

• Human escalation rules for hardship, vulnerability, fraud, identity theft, dispute and legal-status issues.

• Full preservation of recordings, transcripts, disposition codes and decision logs.

• Ongoing review for inaccurate statements, deceptive framing, inappropriate pressure and disparate treatment across languages or borrower segments.

• Procedures for testing voice AI against real-world edge cases before release and continuously monitoring it after deployment.

Ignosis says lenders retain control through their own calling infrastructure and receive recordings, transcripts and outcomes. The company says a lender can stop the calling pipeline if an anomaly is detected or halt contact with a particular borrower in real time through a software connection.

The Larger Industry Signal

Ignosis’ model reflects a growing shift from automation that merely increases call volume to systems intended to make case-level collections decisions. The company says its platform can process tens of thousands of concurrent calls at sub-300-millisecond latency, while using Account Aggregator and GST signals to prioritize outreach and individualize retry timing. That description comes from a June company announcement distributed through BusinessWire India and should be viewed as vendor-provided information.

Still, the underlying proposition is familiar to collection operations leaders: use better data to distinguish willingness from ability to pay, direct routine cases toward efficient digital resolution and position human collectors where their skills carry the most value.

The caveat is that collections is not merely a workflow-optimization problem. It involves people under financial stress, often in emotionally difficult or legally sensitive circumstances. Ignosis’ reported approach—automate the routine work, preserve human intervention for exceptions and audit every conversation—may be a more practical model than an effort to make collections fully autonomous.

The question for the industry is no longer whether AI can place collection calls or send reminders. It is whether agencies and creditors can prove that their AI systems make outreach more accurate, more respectful and more compliant—not simply cheaper or faster.