Borrower calls that confirm identity before saying anything

A reminder call sounds harmless. "Your EMI of a certain amount is due on the fifth." But if the person who picked up is the borrower's brother, their employer, or someone who bought a recycled number, the lender has just disclosed a loan, its amount and its status to a stranger. Human callers make this mistake sometimes. An automated caller, placing calls at volume, can make it systematically.

So the first rule for any automated borrower call is simple: confirm who you are speaking to before you say anything about the account.

Why "Am I speaking to Mr Sharma?" is not enough

Many scripts open by asking for the borrower by name and treating "yes" as confirmation. That checks nothing. Anyone can say yes, and a family member often will, out of politeness or curiosity.

A real check asks for something that the right person knows and that the lender can verify from its own records. Examples include the last four digits of the loan account number, or the borrower's year of birth. The lender supplies the fact; the call compares the answer against the record; only a match opens the conversation.

Designing a good knowledge check

Not every fact makes a good check. A few principles help.

• The lender chooses the fact, per product. What works for a personal loan may not work for a two-wheeler loan sold through a dealer. This belongs in call policy, owned by the lender, not hard-coded in a script.
• Never say the answer first. "Is your account ending in 4-3-2-1?" gives the number away to whoever answers. Ask the question and let the caller supply the answer.
• Allow a limited number of attempts. After that, the call should stop discussing the account, offer a callback or a branch contact, and record that verification failed.
• Handle partial answers carefully. A year of birth that is off by one is still a failed check. Speech recognition errors should lead to a polite retry, not a lenient match.
• Do not reveal why a check failed. "That is not the year we have" tells a stranger they are close. A neutral response is safer.

What the call should never say

Before verification, the call can say who is calling and why in general terms, and nothing more. It should not mention the loan amount, overdue status, EMI, account number or anything that suggests the person owes money.

After verification, it should still speak only from approved material: the lender's call policy and knowledge, and figures read from the loan record. Amounts and dates must come from the system of record, never from a model's guess or a rounded figure. If the call does not have a fact, it should say so and offer to have someone follow up.

There are also things a collections call should never do at all: threaten, shame, contact the borrower's references about the debt, or call at unreasonable hours. RBI's rules on responsible conduct treat calling a borrower before 8 a.m. or after 7 p.m. for recovery of overdue loans as harassment. An automated call made for recovery should be held to at least the same standard as a person.

Saying that the call is automated

Borrowers should know when they are talking to software. Saying so at the start is honest, and it sets expectations about what the call can and cannot do.

The regulators point the same way. The RBI's FREE-AI framework, published in August 2025, says people should be informed when they are interacting with AI systems. TRAI's Telecom Commercial Communications Customer Preference (Third Amendment) Regulations, 2026, notified in September 2026, require senders of automated application-to-person calls to declare that activity and their calling numbers to their telecom provider in advance; undeclared automated calls are treated as unsolicited commercial communication. Separately, TRAI has directed RBI-regulated entities to make service and transactional calls from the 1600 number series, with deadlines in early 2026.

Lenders should take legal advice on how these apply to their own calls. The practical point is that an AI voice programme needs its telecom set-up decided before the first call, not after.

Opt-outs belong to the lender

When a borrower says "do not call me again" or "call me only in Hindi" or "only after six", that preference must be recorded and honoured. It belongs to the lender, per lender, not to the calling platform's shared list. A borrower who opts out of one lender's reminder calls has not opted out of another lender's, and one lender's list should never be visible to another.

Handing over to a person

Some conversations should not stay automated. The borrower asks for a person. They mention hardship, a death in the family, a dispute, or a complaint. They become distressed. Call policy should define these triggers, and when one occurs the call should hand over to a person with the transcript, so the borrower does not have to repeat themselves.

Recording what was agreed

The value of a reminder or verification call lies in what it records. A good call ends by reading back what was agreed: "You will pay the instalment on the twelfth by UPI. Is that right?" Then it stores the answer, any promise to pay with its date, any callback request, and whether verification passed. A person reviewing the account the next day should see exactly what was said and agreed, not a summary of what the model thought happened.

We built Sonics, the voice part of Inferova, around these rules: verification with a lender-supplied fact before any account detail, speech only from approved knowledge and call policy, figures read from the record, read-back confirmation, opt-outs kept per lender, and handover with the transcript. Speech recognition and voices run on the lender's own infrastructure.

If you are planning automated borrower calls and want to see how these checks work, you can request beta access at inferova.com/demo.

Sources