EMI Reminder Calls with AI: Staying Inside RBI Conduct Rules

An AI agent can place EMI reminder calls for banks and NBFCs, but the conduct rules that apply to human collectors apply to it as well. RBI's August 2022 circular on recovery agents bars calling borrowers before 8:00 a.m. or after 7:00 p.m. for recovery of overdue loans, bars persistent calling and harassment, and keeps the regulated entity responsible for its service providers. Design the agent around those limits.

This guide is information, not legal advice. We cite primary or well-established sources where we could verify them, and say so where we relied on a secondary summary. Your compliance and legal teams must decide how the rules apply to your products, borrowers, and contracts, and rules change, so check current text on rbi.org.in before deploying.

What is the difference between a reminder and recovery?

A courtesy reminder before the due date, telling a borrower their EMI is coming up, is different from a call about an overdue account. Many lenders treat the two differently in their policies. RBI's calling-hour rule is written for recovery of overdue loans, so the cleanest and most defensible approach is to apply it to every call your agent makes about a loan, and to keep pre-due and overdue scripts separate.

What does RBI say about calling borrowers?

The Reserve Bank of India's circular RBI/2022-23/108, dated 12 August 2022, 'Outsourcing of Financial Services: Responsibilities of regulated entities employing Recovery Agents', sets out several points relevant to an AI caller. It applies to commercial banks, co-operative banks, NBFCs, asset reconstruction companies, and all-India financial institutions, and excludes microfinance loans covered under separate directions.

• Calling hours: recovery agents must not call borrowers before 8:00 a.m. or after 7:00 p.m. for recovery of overdue loans.
• No persistent calling: the circular names persistently calling the borrower as a form of harassment.
• No intimidation: regulated entities must ensure that they and their agents do not use intimidation or harassment of any kind, verbal or physical.
• Outsourcing liability: ultimate responsibility for outsourced activities remains with the regulated entity, which is responsible for the actions of its service providers, including recovery agents.

The practical consequence for an AI agent vendor relationship is that outsourcing the calls does not outsource the responsibility. The bank or NBFC must be able to show how the agent is configured, supervised, and monitored.

What does the Fair Practices Code add?

RBI's Fair Practices Code for NBFCs, published on rbi.org.in, includes a requirement that in recovery of loans, NBFCs should not resort to undue harassment, such as persistently bothering borrowers at odd hours or using muscle power, and that staff be adequately trained to deal with customers appropriately. It does not itself state a numeric calling window. The 8:00 a.m. to 7:00 p.m. window comes from the 2022 circular above.

Read 'staff adequately trained' as applying to anyone, or anything, speaking for the lender. For an AI agent, that translates into reviewed scripts, tested behaviour on difficult calls, and routine call sampling.

What do the Digital Lending Directions add?

The Reserve Bank of India (Digital Lending) Directions, 2025, issued on 8 May 2025, consolidated earlier digital lending guidance. Law-firm summaries of the Directions (for example, Legal500 and Lexology, which we used as secondary sources) report these points relevant to collections communication: details of any recovery agent must be communicated to the borrower by SMS or email before the agent contacts them, and regulated entities must seek explicit consent before collecting borrower data and limit collection to what is needed.

These are secondary summaries. Before relying on any specific clause, read the Directions on rbi.org.in and have counsel confirm how they apply to your lending model, including whether an AI calling vendor counts as a service provider or recovery agent in your arrangement.

How does the DPDP Act interact?

India's Digital Personal Data Protection Act, 2023 and the Digital Personal Data Protection Rules, 2025 (notified on 14 November 2025, with obligations phased in over roughly eighteen months, as described in the Press Information Bureau's release) govern how personal data is processed. A reminder call uses personal data: a name, a phone number, loan details, and often a recording and transcript.

• Have a clear lawful basis and notice for contacting the borrower and recording the call.
• Collect and keep only what the reminder needs; do not read out more account detail than necessary.
• Define retention for recordings and transcripts and delete them on schedule.
• Be ready to honour access, correction, and withdrawal requests.
• Check what your vendor does with call data, including where it is stored and whether it is used for anything beyond your service.

What should an AI reminder agent say?

• Identify the caller and the lender at the start: the agent's name or role, that it is an automated voice assistant, and the lender it calls on behalf of.
• Verify before disclosing: confirm they are speaking with the borrower, using an approved method, before mentioning loan details. If someone else answers, say nothing about the loan and offer to call back.
• State the reminder plainly: the EMI amount and due date, from the lender's system of record, nothing improvised.
• Offer helpful paths: payment methods, a payment link by an approved channel, or a connection to a person if the borrower has a difficulty.
• Record the outcome: paid, will pay by a date, needs help, disputes the amount, or asks not to be called.
• Close courteously, and confirm any promise the borrower made, as their own words.

What must it never do?

• Call outside the permitted window, or call repeatedly in a way that could be read as persistent.
• Threaten, shame, or imply legal, police, or social consequences it has no authority to impose.
• Disclose loan details to anyone other than the verified borrower, including family members, colleagues, or whoever picks up.
• Invent figures: late fees, interest, settlement amounts, or concessions that are not in the system of record or approved scripts.
• Give legal or financial advice, negotiate settlements, or promise waivers.
• Continue after a borrower says they do not want to be contacted, disputes the debt, or becomes distressed.
• Contact third parties from the borrower's contacts or phone directory.

What does a compliant call look like in practice?

The outline below is a structure to adapt with your compliance team, not approved wording. It shows the order of steps: disclose, verify, state the fact, offer options, record the outcome. Each line should come from approved script text and system-of-record data, never from the model's improvisation.

1. 01Opening: 'Namaste, this is an automated voice assistant calling on behalf of [lender name]. Am I speaking with [borrower name]?'
2. 02If the person is not the borrower: 'Thank you, I will try again later,' and end the call without mentioning the loan.
3. 03After verification: 'Your EMI of [amount] for your loan is due on [date].'
4. 04Options: 'You can pay through the app, a payment link we can send, or I can connect you to a team member.'
5. 05If the borrower reports a difficulty: 'I am sorry to hear that. Let me connect you to someone who can help,' followed by escalation with a summary.
6. 06Closing: repeat any commitment the borrower made, thank them, and end the call.

Test the script against hostile and confused callers before launch: someone who denies owning the loan, someone who hangs up and calls back, someone who speaks only in a regional language, and someone who asks whether they are speaking with a person. Each should end in a safe outcome, whether that is a clean exit, a callback, or a handover to a human.

How should escalation and data handling work?

Escalation is the safety net. Define triggers in advance and test them. Typical triggers are: the borrower mentions hardship, illness, bereavement, or job loss; disputes the amount or says the loan is not theirs; is abusive or distressed; asks for a person; or the agent cannot verify identity. On escalation the agent should hand over a short summary so the borrower does not repeat everything.

• Log every call attempt with time, number, outcome, and script version, so you can show compliance with calling hours and attempt limits.
• Keep recordings and transcripts under a defined retention schedule, with access limited by role.
• Run a weekly quality sample, with extra weight on calls where the borrower sounded unhappy.
• Keep a grievance route: tell borrowers how to complain to the lender, and be aware that borrowers may take complaints to the RBI Integrated Ombudsman Scheme.
• Review scripts with compliance whenever product terms or RBI directions change.

When not to use an AI agent for EMI calls

• Accounts in dispute, in hardship programmes, or under legal process: use trained humans.
• Late-stage overdue accounts where tone and judgment matter most.
• Borrowers who have said they prefer written communication.
• Any segment where you cannot yet verify identity reliably over the phone.

For the wider picture of loan-related calls, including intake and status checks, see our guide to AI voice agents for loan applications and financial services.

Sources

• Reserve Bank of India, circular RBI/2022-23/108, 'Outsourcing of Financial Services - Responsibilities of regulated entities employing Recovery Agents', 12 August 2022: rbidocs.rbi.org.in (confirmed through the TaxGuru reproduction of the circular).
• Reserve Bank of India, Guidelines on Fair Practices Code for NBFCs: rbi.org.in/commonman/english/Scripts/Notification.aspx?Id=1572
• Reserve Bank of India (Digital Lending) Directions, 2025 (8 May 2025), as summarised by Legal500 and Lexology; read the primary text on rbi.org.in before relying on specific clauses.
• Press Information Bureau, Government of India, 'DPDP Rules, 2025 Notified', November 2025: static.pib.gov.in