For a collection agency, an AI voice agent does the repetitive, high-volume calls: pre-due reminders, early-bucket follow-ups and broken-promise calls, in the borrower's language, within RBI's permitted hours. It captures a promise to pay with a date, sends a payment link during the call, and records everything for audit. Your trained agents then work the harder cases: disputes, hardship and settlements. On Edesy this runs as an AI debt collector and costs Rs 4-6 per answered minute, with unanswered calls free.
This guide is for third-party collection agencies (DCAs) working bank and NBFC portfolios: what to automate, bucket-wise scripts, compliance and cost per connected call.
Test on one bucket: Start free on Edesy with a small allocation, or talk to us about portfolio-scale calling.
Why agencies are adding AI calling
A DCA's economics depend on contacts per agent per day. Much of an agent's shift goes to numbers that don't answer, repeat reminders, and logging outcomes. Lenders also push for more calls in the early buckets, where recovery is cheapest.
AI changes the mix: the AI makes every early-bucket call and every reminder, and your team's time goes to conversations that need a person.
What to automate by bucket
| Bucket | AI's job | Hand to a person when |
|---|---|---|
| Pre-due (3-5 days before) | Friendly reminder, payment link | Borrower raises a dispute |
| 1-30 DPD | Reminder, reason for delay, promise to pay | Hardship, restructuring request |
| Broken promise | Follow-up on the missed date, new date | Second broken promise |
| 31-60 DPD | Reminder and contact confirmation | Most conversations; AI prepares context |
| 60+ DPD, settlements, disputes | Not recommended | Always |
Bucket-wise scripts (Hinglish)
Pre-due reminder
Namaste {{name}} ji, main {{lender}} ki taraf se ek automated call kar rahi hoon. Aapki {{amount}} rupaye ki EMI {{due_date}} ko due hai. Kya aap tab tak payment kar payenge? Main aapko payment link SMS par bhej sakti hoon.
1-30 days past due
Namaste {{name}} ji, main {{lender}} ki taraf se ek automated call kar rahi hoon. Aapki {{amount}} rupaye ki EMI {{due_date}} se pending hai. Kya aap bata sakte hain aap kab tak payment kar payenge?
Broken promise
Namaste {{name}} ji, aapne {{promise_date}} tak payment ka vaada kiya tha, jo abhi tak nahi hua hai. Kya sab theek hai? Aap nayi tareekh bata sakte hain?
Keep every script polite and factual. Never threaten, never mention consequences the lender has not approved, and never discuss the debt with anyone except the borrower. More templates: Hindi EMI collection voice agent.
What the agent captures
Define these as data fields before the campaign:
- Outcome: will pay, already paid, needs time, dispute, wrong number, refused
- Promised date and promised amount
- Reason for delay (fixed options such as salary delay, medical, business loss)
- Escalation needed (yes/no)
Edesy extracts these fields after each call. You get a clean promise-to-pay list by date, plus the recording and transcript for every call, which is what lenders ask for in audits. Export to CSV, Google Sheets, a webhook or your collection system.
Compliance: RBI rules for recovery calls
The lender stays responsible for its agents' conduct, including yours. The key rules from RBI's August 2022 circular on regulated entities employing recovery agents:
- Calling hours: no calls to borrowers before 8:00 am or after 7:00 pm (Business Standard).
- No intimidation or harassment, verbal or otherwise, of borrowers or their family and friends.
- No inappropriate messages or threatening calls.
How to build this into an AI campaign:
- Set the campaign calling window inside 08:00-19:00 IST. Edesy pauses calls outside the window.
- Get the scripts approved by the lender before launch.
- Record every call and keep transcripts for the retention period the lender requires.
- Identify the caller and say it is an automated call.
- Never call third parties about the debt; handle "wrong number" by ending politely.
For a deeper read, see RBI guidelines for AI collection calls and RBI rules on debt collection automation. Also follow DPDP rules for personal data: DPDP compliance for voice AI.
Languages
Borrowers answer more readily in their own language. Edesy agents speak Hindi, Hinglish, Marathi, Bengali, Telugu, Tamil, Gujarati, Kannada, Malayalam, Odia, Assamese and other Indian languages, plus English. Split the allocation file by state or language column and run one campaign per language.
Cost per connected call
On Edesy, calls are billed per started minute at Rs 4-6 per minute depending on your prepaid top-up tier, and unanswered calls are not charged. There are no seat or licence fees.
A worked example with assumptions to replace:
- Allocation of 20,000 accounts, 55% answered across retries = 11,000 answered calls
- Most reminders finish within a minute = 1 billed minute each
- 11,000 x Rs 4-6 = Rs 44,000-66,000, or Rs 4-6 per connected call
Compare that with your cost per connected call today (agent cost per month divided by connected calls per agent per month). For large portfolios, ask us about volume and per-second pricing.
How agencies roll it out
- One lender, one bucket. Start with pre-due or 1-30 DPD on a single allocation.
- Pilot 500 accounts. Listen to recordings, check promise capture accuracy, adjust the script.
- Compare against a matched group called by your team: contact rate, promises, and payments.
- Scale to other buckets and lenders once the numbers hold.
Also see collections for NBFCs and microfinance and how to automate debt collection in India.
Start with one allocation
Create a free account, upload a small allocation and run your first reminder campaign inside RBI hours. Working large portfolios across lenders? Talk to us about volume pricing and setup.