Collections is the use case where automated calling gets the most interest and deserves the most caution. The volume argument is obvious — a delinquent book is thousands of accounts and a tele-calling floor that cannot cover them. The risk argument is equally obvious, and it is not really about technology.
An automated caller is held to exactly the same standards as a human one. Nothing about a machine placing the call relaxes any rule, and the borrower-facing conduct expectations in the RBI's Fair Practices Code apply to your agent the way they apply to your staff.
This is a practical summary, not legal advice. Your compliance team should approve any collections prompt before it dials a single account.
The boundaries that are not negotiable
- Calling hours. Recovery contact is expected between 8 AM and 7 PM. Configure the window in the platform and do not make exceptions for "just this campaign".
- No harassment. Repeated calling to pressure a borrower is the thing regulators act on. Frequency caps are not a nicety; they are the control that keeps you defensible.
- No third parties. Do not discuss the debt with a spouse, employer, neighbour or anyone who is not the borrower. If someone else answers, you leave a callback request and nothing else.
- No threats, no false urgency. No implied legal action that is not actually in motion, no invented consequences, no "final notice" that is not final.
- No misrepresentation. The agent identifies the lender and the fact that it is a call about a payment. It does not pretend to be a law firm, a court or a bank official.
- Grievance route. A borrower who wants to complain gets told how.
Every one of these is easier to guarantee with an automated caller than a human one, which is the genuinely underrated argument for it. A prompt does what it is told on call five thousand exactly as on call one, it never loses its temper at 6 PM on a Friday, and every word of every call is transcribed and reviewable.
Where automation fits, by bucket
| Stage | Automated call | Why |
|---|---|---|
| Pre-due | Yes, ideal | Pure reminder, no dispute possible |
| 1–15 days | Yes | Mostly oversight; a nudge resolves it |
| 16–30 days | Yes, with fast handoff | Reasons start to matter; capture and escalate |
| 30–90 days | Triage only | Establish contact and intent, then hand to a human |
| 90+ days | No | Restructuring and settlement need authority and judgement |
The rule underneath the table: an automated agent may remind, ask and record. It may not negotiate, threaten or settle.
The frequency cap is the control that matters
One contact attempt per day at most, and in practice fewer. A reasonable pattern is day 3, day 8 and day 15 after the due date, then a human.
Three calls in a day to the same number is harassment whether a person or a system placed them, and an autodialler makes that mistake effortlessly. Set the cap in the platform rather than trusting the campaign design, because campaigns get duplicated and re-run.
Why the transcript is your best defence
The usual complaint in Indian collections is that an agent was abusive, threatening or contacted a family member. With human tele-calling that becomes one person's word against another's, and the lender generally loses.
With an automated caller you have the exact prompt, a full transcript of every call, and a timestamp. If the prompt forbids threats and the transcript shows none, the dispute resolves quickly. That evidentiary position is a stronger argument for automating collections than the cost saving.
What it demands in return is that you actually read them. Reading transcripts systematically applies here more than anywhere.
The uncomfortable recommendation
If you run a collections operation, start with the pre-due reminder rather than the delinquent book. It carries almost no regulatory risk, it is cheaper per account, and it shrinks the population that reaches collections at all.
It is also the campaign where your team learns how the system behaves before it is pointed at borrowers who are already under stress. The lending page sets out the boundaries in full.