Voice AI5 min read

Voice AI for NBFC Collections in India: RBI and TRAI Rules

Voice AI for NBFC collections in India must follow RBI recovery rules and TRAI number series. Here is how to build a compliant AI collections agent.

HM
Harshit Makraria

Voice AI for NBFC collections in India has to follow the same conduct rules as a human recovery agent: calls mostly between 8 am and 7 pm, recorded conversations, no harassment, and the right TRAI number series. The regulator does not exempt automation, so a compliant AI collections agent is one with its calling window, recording and scripts built in from day one.

Key takeaways

  • RBI's new recovery directions reportedly take effect on January 1, 2027, and they limit recovery calls to 8 am to 7 pm and require recording.
  • Collections calls to existing customers belong on the TRAI 1600 series, not the 140 promotional series.
  • Build compliance into the system (time gates, recording, scripts, opt-out), not into a policy document.

What do the new RBI recovery rules require?

According to Free Press Journal, the RBI guidelines take effect on January 1, 2027 and cover banks and regulated lending institutions. Recovery calls and visits are generally permitted only between 8 am and 7 pm, and contact outside that window needs the borrower's prior request or approval.

The same report says lenders must record recovery conversations, keep records of the number and timing of calls, and retain them for at least six months. Lenders also stay accountable for outsourced agents, and threats and public shaming are banned. I found no rule written specifically for AI voice calls, so assume the human-agent standard applies to your bot. Dates and details have varied across news coverage, so confirm the final text on rbi.org.in before you finalise your design.

Which TRAI number series should collection calls use?

TRAI has set aside two series. Per The Tribune, the 1600 series is mandated for service and transaction calls from RBI, SEBI, IRDAI and PFRDA regulated entities to their existing customers, and these calls cannot be tagged, filtered or blocked. The 140 series is for promotional calls, and customers can block it by sector through the DND registry.

A reminder about an EMI is a service call to an existing customer. A cross-sell offer to the same person is promotional. Keep the two on separate numbers and separate campaigns, or a DND complaint on the promotional side can drag your whole calling operation into review.

How to build a compliant AI collections agent

ControlWhat to buildWhy it matters
Calling windowHard time gate at the dialer, 8 am to 7 pm in the borrower's timezone, no override by campaign managersCalls outside the window need borrower approval
Recording and logsStore audio, transcript, number and timestamp per attempt for at least six monthsYou must show conduct, not just policy
Script guardrailsApproved script paths, no threats, no third-party disclosure, no pressure on relativesHarassment rules apply to bots too
Identity checkConfirm you are speaking to the borrower before stating any amountPrevents disclosing debt to the wrong person
Human handoffTransfer to a person on dispute, distress or requestSensitive situations need judgment

The pattern is the same one used in the AI debt collection compliance guide and the voice AI governance and platform selection post: rules live in code, so an agent cannot talk its way past them.

Where does AI help most in collections?

The highest value is in the early buckets: friendly EMI reminders, payment-link delivery, promise-to-pay capture and callback scheduling. These are high volume and low emotion, which is where automation is safest. Late-stage and disputed accounts should route to people. Our own live collections system, built to US TCPA standards, has handled about $48.9M in customer accounts across 5,386 call attempts with a 17% productive contact rate, and the lesson carried over: tight rules and clean handoffs matter more than clever dialogue.

Design the handoff early. A warm transfer with the call summary already on screen avoids making a stressed borrower repeat themselves, as covered in our warm transfer guide. For the dialing and recording stack itself, see our AI calling system.

What should you do before the January 2027 date?

Start with an audit of the current calling stack. List every campaign that touches a borrower, the number each one dials from, the hours it runs and where the recordings go. Most gaps show up here, usually a legacy dialer with no hard time gate or a recording tool that keeps audio for a few weeks only.

Then pilot the AI agent on one early-bucket portfolio with a human reviewing a sample of transcripts every week. Track promise-to-pay rate, complaint count and transfer rate, not just connect rate. A pilot that keeps complaints flat while lifting promise-to-pay is the evidence your compliance team will want before you scale. Keep every prompt change under version control so you can show exactly which script was live on any given day.

Common mistakes to avoid

  • Using one number and one campaign for both reminders and offers.
  • Treating the calling window as a setting instead of an enforced gate.
  • Letting the model improvise wording on amounts, penalties or consequences.
  • Skipping recording retention, then being unable to answer a complaint.

Frequently asked questions

Can an AI voice agent legally make loan recovery calls in India?

I found no rule that bans it, and recovery conduct rules appear to apply to automated calls the same way as to human agents. Check the final RBI directions and take legal advice for your lender type.

What calling hours apply to NBFC collection calls?

Reported RBI directions generally allow recovery calls between 8 am and 7 pm, with calls outside that window needing the borrower's prior request or approval. Verify against the final RBI text.

Which number series should a lender use for EMI reminders?

TRAI mandates the 1600 series for service and transaction calls from regulated financial entities to existing customers. Promotional calls belong on the 140 series.

If you want this built for your business, book a free audit with Nexica AI. We build production grade AI systems in 14 days.

Voice AIAI AutomationNexica AI
Want this built for your business?
HM
Written by Harshit Makraria, founder of Nexica AI

Harshit builds production voice AI, workflow automation and AI agent systems for operators. Nexica AI runs a live voice AI collections system that has handled $48.9M in customer accounts, and has delivered 100+ systems in 14 day builds.

Free AI Audit