Voicebot for NBFC Collections vs Human Agents: Which Calls to Automate Before 2027

A voicebot for NBFC collections is an AI-driven calling system that speaks with borrowers in their language, confirms dues, records promises to pay and logs every outcome, so human collectors only take the calls that need judgment.

Most NBFC collection heads are no longer asking whether to use a voicebot. They are asking where it belongs. Put it on the wrong accounts and complaints rise; keep it too narrow and your agents spend their day reading out EMI amounts. The question has also become sharper because of RBI's new recovery directions for NBFCs, which take effect on 1 January 2027 and turn call hours, call frequency and call recording into things you have to prove, not just promise. This guide gives a bucket-by-bucket split between voicebot and human agents, the handoff rules between them, and a 45-day plan to put it in place before the new rules start.

Should an NBFC use an AI voicebot or human agents for collection calls?

Use both, divided by delinquency stage and by what the call has to achieve. A voicebot should handle high-volume, low-judgment calls such as pre-due reminders and early-bucket follow-ups. Human agents should handle negotiation, disputes, hardship and anything near write-off.

Early in the cycle most borrowers have forgotten, been short for a few days, or need a payment link. Those conversations follow a script, and a voicebot can hold hundreds of them at once in Hindi, English or a regional language. Later, the borrower is avoiding calls, disputing charges or in real difficulty, and needs someone who can read the situation and offer options within policy. An AI voicebot for NBFC collections is a capacity tool for the first kind of call, not a replacement for the second.

Which DPD buckets suit a voicebot, and which need a person?

Days past due (DPD) is the cleanest way to divide the work, because it already drives your provisioning, your reports and your escalation matrix. The split below is a starting point; adjust it to your product mix and ticket size.

Two adjustments come up often. For small-ticket digital loans, the voicebot can run up to 60 DPD because there is little to negotiate. For secured loans with larger balances, many teams pull agents in from 15 DPD because one well-handled call is worth more than ten reminders. Whichever split you pick, write it into the recovery policy, since the new directions expect a documented escalation matrix.

What do RBI's January 2027 recovery rules change for collection calls?

On 6 August 2026 RBI issued the Non-Banking Financial Companies (Responsible Business Conduct) Third Amendment Directions, 2026, with parallel directions for banks and housing finance companies. From 1 January 2027, NBFCs need a board-approved recovery policy, recorded recovery calls kept for six months, contact only between 8 am and 7 pm, and grievance officer details on every recovery communication.

The directions are written around employees and recovery agents. A voicebot calling on the NBFC's behalf is doing the same job, so the safe reading is to hold it to the same standard. Confirm that reading with your compliance team, but design for it now rather than retrofitting later. Strictly, the recovery section applies to borrowers in default, so pre-due reminders sit outside it; most teams still apply the same hours and recording rules to every collections call because it keeps one set of controls.

The calling-number point is not new. TRAI's November 2025 direction moved RBI-regulated entities onto the 1600 series for service and transactional calls, with NBFC deadlines that passed in early 2026; our 1600-series deadline explainer lists the dates by entity type. For the wider compliance picture, including Fair Practices Code history and the FREE-AI report, see our NBFC RBI and TRAI compliance guide.

How do voicebot calls and agent calls compare?

A voicebot wins on reach, consistency and record-keeping. An agent wins on judgment, negotiation and reading distress. Cost per call favours the bot by a wide margin, but cost per rupee recovered depends on putting each call type in the right hands.

The two risk rows are why the split matters. A bot used beyond its stage creates complaints through rigidity; an agent team stretched across every bucket creates complaints through fatigue and targets. Recording both, and scoring the agent calls with AI Voice Analysis, lets you see which problem you actually have.

When should a voicebot hand a borrower to a human agent?

Hand over the moment the conversation leaves the script: a dispute, a hardship signal, a request for options the bot cannot offer, or a second broken promise. A good handoff passes full context so the borrower never repeats themselves.

Triggers that should end the bot conversation

1. The borrower disputes the amount, a charge or a payment already made.
2. The borrower mentions job loss, illness, a death in the family, a wedding or any emergency.
3. The borrower asks about restructuring, a moratorium or settlement.
4. This is the second broken promise to pay on the account.
5. The borrower is angry, asks for a person, or asks who the caller is twice.
6. Identity cannot be verified after two attempts, or a third party answers.
7. The borrower says they did not take the loan, which may indicate fraud.

What the agent should see when the call lands

The agent's screen should show the bot's transcript, the reason for handoff, the DPD and amount, past promises and whether they were kept, the borrower's preferred language and time slot, and any flags such as a recorded bereavement. If the handoff happens outside agent hours or after 19:00, the bot should book a callback inside the window instead of transferring.

Does the borrower need to be told they are speaking to AI?

Tell them. RBI's FREE-AI report of August 2025 recommends that regulated entities disclose when customers are dealing with AI and give them a way to reach a human. The report is advisory today, but it signals where supervision is heading, and the January 2027 directions already require you to say the call is recorded.

In practice this costs one sentence. An opening line such as "This is an automated call from [NBFC name] about your loan account; the call is recorded" covers both points. The recordings themselves are personal data, so retention, access and deletion should also follow your plan under the Digital Personal Data Protection Act and its 2025 Rules.

How do you roll out a voicebot and agent split? A 45-day plan

1. Days 1 to 7: write the split. Agree the bucket table above with collections, risk and compliance. Add the handoff triggers and frequency caps to the recovery policy draft, since the board-approved policy has to cover the escalation matrix anyway.
2. Days 8 to 15: build the scripts. One script per stage, in each language you need. Include identity verification, the recording and AI disclosure, approved consequence wording, the grievance officer details for follow-up messages, and the exact handoff lines. Have legal review the consequence wording.
3. Days 16 to 22: connect the systems. Push daily allocation files from the loan management system, sync outcomes and promises back after every call, and set the 08:00 to 19:00 hard stop and per-account caps in the scheduler. Confirm all outbound traffic uses your 1600-series numbers.
4. Days 23 to 35: pilot on one bucket. Start with pre-due reminders for one product and region. Listen to a sample of bot calls every day for the first week, especially handoffs and wrong-party calls. Fix script lines that confuse borrowers.
5. Days 36 to 45: extend and measure. Add 1 to 30 DPD, move agents off reminder calls and onto 31 DPD plus, and start the weekly scorecard in the next section.

Which numbers show whether the split is working?

Track contact, promise and roll-forward rates by bucket, and track compliance numbers that should be zero. If early-bucket roll rates fall while complaints stay flat or drop, the split is working.

What results have NBFCs reported?

Two Elision deployments show what the voicebot side of the split can do. Dvara KGFS, which lends to rural households, runs Elision's AI Voicebot in Hindi, English, Kannada, Telugu and Malayalam. Elision reports that the voicebot now carries 45% of Dvara KGFS's calling volume, contactability rose from 52% to 58%, promise-to-pay generation grew by 24%, and collection conversion went from 31% to 42%. The full write-up is in the Dvara KGFS success story.

At Capri Loans, manual EMI reminder calls took three to four minutes each. Moving them to the voicebot saved more than 400 agent-hours a month, according to Elision's Capri Loans case study, with calls placed on 1600-series numbers. On the agent side, collection agency Astute reported an 82.57% increase in collections after moving to DialShree's Collection Module (Astute success story).

If you are still choosing a platform rather than designing the split, the debt collection software buyer's guide covers what to check in a demo, our auto dialer for NBFC collections page explains the agent-side dialing, and the overview of AI automation in NBFC recovery covers the wider stack.

Glossary

DPD (days past due)

Days since an instalment fell due without payment; the basis for collection buckets.

Roll rate

The share of accounts that move from one DPD bucket to the next in a period.

Right-party contact (RPC)

A call answered by the borrower or guarantor, as opposed to a family member or wrong number.

Promise to pay (PTP)

A borrower's commitment to pay a stated amount by a stated date; see the PTP glossary entry.

Handoff

Transfer of a borrower from the voicebot to a human agent, with the conversation context attached.

1600-series number

The number range TRAI requires RBI-regulated entities to use for service and transactional calls.

Escalation matrix

The documented sequence of recovery actions by stage, which RBI expects in an NBFC's recovery policy.

Frequently asked questions

Sources and regulatory references

• Reserve Bank of India, Non-Banking Financial Companies (Responsible Business Conduct) Third Amendment Directions, 2026, RBI/2026-27/230, 6 August 2026: rbi.org.in
• Reserve Bank of India press release, Amendment Directions on conduct in recovery of loans and engagement of recovery agents, 6 August 2026: rbi.org.in
• Reserve Bank of India, Report of the Committee on the Framework for Responsible and Ethical Enablement of AI (FREE-AI), 13 August 2025: rbi.org.in
• TRAI Press Release No. 135/2025, direction on phase-wise adoption of the 1600 series by BFSI entities, 19 November 2025: trai.gov.in