The short answer: a collection voicebot earns its keep in the early buckets. DPD 1–30 (days-past-due) is where 25–35% of delinquent accounts self-cure if someone simply calls them — politely, on the right number, at the right hour, with a payment link ready. A voicebot does this at ₹2–3 per connected minute at unlimited concurrency, which changes collections math completely: you can call every delinquent account every day instead of triaging. This playbook lays out strategy, tone, scripts and KPIs for each bucket, so a lender can deploy the DPD 1–30 bucket in week one, and DPD 31–60 in week two.
What DPD buckets mean (and why bots fit the early ones)
| Bucket | Reality of the borrower | Best channel strategy |
|---|---|---|
| DPD 1–30 | Forgetfulness, cash-flow timing, side-thought about default | Voicebot-first, daily, payment links live |
| DPD 31–60 | Structural strain; some genuine hardship | Voicebot + scheduled human follow-up on hardships |
| DPD 61–90 | Complex hardship, partial-payment negotiators, some disputes | Human-first, voicebot used for right-party-contact salvage |
| DPD 90+ | NPA territory, field agency work | Field/human; voicebot used only for audit-ready contact attempts |
The industry loses money where volume is high but leverage is low — exactly DPD 1–30. Field agents are rightly reserved for buckets where face-to-face leverage matters. The voicebot fills the early-bucket gap: thousands of soft reminders, with zero aggression and a full audit trail on every dial.
DPD 1–30: the soften-first strategy
- Tone — helper, not collector. The position is: "your EMI slipped, life happens, here is the link, three taps and you are current." Nothing about consequences, nothing about the loan amount growing, no urgency theatre.
- Cadence — every 2 days, 3 touches per week max, 10 AM–7 PM. Same-day bounce calls (Script 6 in our loan reminder voicebot guide) have the single highest save-rate in the whole reminder+collection stack.
- Offer on-call payment links — UPI link by WhatsApp is the close; the call ends the moment the borrower taps it. Friction is the killer.
- Hardship keywords route out instantly — job loss, medical, dispute, or "I cannot pay this month" without a date goes to a human, not to a second bot personality.
Target allocation of effort: DPD 1–30 volume is 3–5× the rest of the book combined. Automating here alone moves the NPA curve; buckets stay human until this bucket is bot-native.
DPD 31–60: the structured-repayment strategy
- Tone — respectful, firmer, options-driven. The borrower is now flagged internally; the script introduces the overdue amount factually, one restructure option ("mini-schedule: two EMIs of ₹4,200 over 15 days"), and asks for a commitment date like a human collector would.
- Cadence — alternate days, and schedule human callbacks for anyone who signals partial-payment intent.
- PTP discipline tightens: one PTP per borrower per cycle, auto-broken if the promised date passes (broken-PTP call follows Script 5).
- Compliance elevates: all collections-conduct rules apply strictly now — 8 AM–7 PM window, no third-party disclosure, call recording for audit; the standard TRAI boundaries remain as covered in the TRAI and DND guide.
DPD 61–90: escalate the conversation, not the tone
- Human-led. The voicebot is used for two jobs only: (1) right-party-contact setting — confirming reachability, best time, willingness — which arms the human officer, and (2) unsold-contact salvage at scale, logging every attempt in audit-ready form so the file shows continuous good-faith outreach.
- No negotiation by bot, ever. Settlement, waiver, restructuring conversations belong to licensed humans with writing authority.
- Hardship has primacy: any distress signal exits automation immediately — RBI fair-practice conduct applies without exception at this depth.
DPD 90+: NPA-track, agency prep
- Collections move to field agencies and specialized desk teams. The voicebot does not call borrowers here.
- It runs the documentation trail — padded contact-attempt summaries of every previous bucket, accessible in one audit export, which is what prevents later "we were never contacted" disputes from stalling legal action.
The KPIs that prove it is working
| Metric | DPD 1–30 target | What it measures |
|---|---|---|
| Right-Party-Contact (RPC) rate | 55–75% | % of dial outcomes that reached the borrower, not family or voicemail |
| PTP rate | 30–50% | % of RPC calls that ended with a commitment date |
| PTP kept rate | 60–75% | % of promises that became payments — the real quality number |
| Cure rate (bucket exit) | 25–35% | % of DPD 1–30 accounts that returned to current within the cycle |
| Cost per contact | ₹4–9 | Voicebot economics vs. ₹80–200 for a human-collector dial |
| Complaint rate | <0.1% | Escalations to the lender or ombudsman — the tone check |
One number matters more than the rest in week one: cure rate. If a polite daily-call voicebot on DPD 1–30 cures even 20% of the bucket, the rest of the automation case funds itself. Everything else — RPC, PTP — is diagnostics on the way there.
Hand-off rules that keep bots and humans in their lanes
- Hardship → human (job loss, medical, bereavement signals) — instantly, with context notes attached to the LMS record.
- Dispute → human ("I never took this loan", fraud claims) — never argued with the bot.
- Explicit refusal or opt-out → internal do-not-call list, honored on the next dial and visible to every future campaign.
- Partial-payment negotiation → human, because only humans can offer written restructures.
- Everything else stays with the bot — paid confirmations, promise dates, payment links, appointment-style callbacks.
Where to start in your book
- Pick the DPD 1–30 island: one product line (personal loan is ideal — high volume, clear transactionality).
- Import the delinquent list with due-date metadata; verify NACH-mandate status per row.
- Run the three-message cadence (reminder → link → broken-PTP) for one full cycle.
- Watch the RPC and cure numbers; expand to DPD 31–60 with the structured-repayment script only when DPD 1–30 numbers trend at the targets above.
The full page with the live call-script preview, pricing and LMS integration detail is here: Voicebot for NBFC — Loan Reminders, EMI Alerts & Collection Calls. And the reminder side of the stack — preventing accounts from entering these buckets at all — is in the loan reminder voicebot scripts guide.