How Voice AI Manages Bucket-2 and Bucket-3 Collections Effectively

In the language of Indian banking collections, "buckets" define the depth of a borrower's delinquency by Days Past Due (DPD). Bucket-0 and Bucket-1 (DPD 0–30) are early interventions where most delinquencies self-cure with a simple reminder. Bucket-2 (DPD 31–60) and Bucket-3 (DPD 61–90) are the critical battleground — where the probability of NPA formation rises sharply, where borrowers need both pressure and solutions, and where the complexity of each case demands more sophisticated engagement.

Voice AI for Bucket-2 and Bucket-3 collections is not just about making more calls. It is about making smarter calls — collecting the right information, presenting the right options, executing SARFAESI-safe language, and ensuring that every genuinely distressed borrower reaches a human who can help them restructure, while every strategic defaulter faces clear, documented consequences.

This guide breaks down the Bucket-2 and Bucket-3 AI collections methodology, the compliance requirements, and the outcomes Indian lenders are achieving.

Why Bucket-2 and Bucket-3 Are Different from Bucket-1

The difference between Bucket-1 and Bucket-2/3 is not just time — it is intent and complexity:

Bucket-1 characteristics:

• Usually accidental (auto-debit failure, forgot payment)
• Borrower is cooperative and responsive
• Simple reminder + payment link resolves most cases
• Low handling complexity

Bucket-2/3 characteristics:

• Mix of genuine financial stress, cash flow disruption, deliberate avoidance, and dispute
• Borrower may be defensive, anxious, or deliberately uncontactable
• Simple reminder does not resolve; structured negotiation required
• Credit bureau impact is now live (30+ DPD is reported to CIBIL, Experian, Equifax, CRIF)
• For secured loans, SARFAESI notices may be approaching
• Restructuring or settlement may be necessary

The AI handling Bucket-2 and Bucket-3 must be significantly more sophisticated than a Bucket-1 reminder agent — and the stakes of getting it wrong are much higher.

The Bucket-2 AI Collections Playbook (DPD 31–60)

Objective

Convert overdue accounts to "Promise to Pay (PTP)" with a committed date, or qualify genuinely distressed accounts for restructuring/hardship programme, or escalate strategic defaulters to human collections.

Call Architecture

Call 1 (DPD 32–35): Structured Inquiry

The first Bucket-2 call should acknowledge the pattern (two missed payments or one significant overdue) while remaining solution-oriented:

1. Authenticate borrower
2. State the overdue amount and months clearly: "Your account is overdue by ₹X for [month(s)]. We want to help you get this sorted."
3. Open-ended question: "Can you help me understand what's happened? Has there been a change in your financial situation?"
4. Based on response, navigate one of three paths:

Path A — Temporary cash flow issue:

• "That makes sense. How long before you can make a payment? Could you manage ₹X by [date]?"
• Set PTP with specific date and amount
• Send SMS confirmation of PTP
• Follow up on PTP date

Path B — Genuine financial distress:

• "I understand this is difficult. We have a few options that may help — a restructuring programme or an EMI break. Can I connect you with our restructuring team?"
• Create hardship flag, schedule human callback within 24 hours
• Do NOT continue collections script

Path C — Dispute:

• "You're saying this charge is disputed? Let me create a ticket for you right now — ticket number is XXXXX. Collections on this amount will be held until the dispute is resolved."
• Route to dispute team

Path D — No engagement / avoidance:

• Leave structured message: "This is [Lender] regarding your overdue account. The outstanding is ₹X. To avoid credit bureau reporting, please call us back at [number] or pay at [link]."
• Schedule follow-up call in 48 hours

Call 2 (DPD 40–45): PTP Follow-up or Escalation

If a PTP was set, the follow-up call either:

• Confirms payment received: "We received your payment of ₹X on [date]. Your account is now current. Thank you."
• Follows up on missed PTP: "We had a commitment for ₹X by [date]. We haven't received it. Can you confirm when you'll be able to pay?"
• Escalates broken PTP to human collector

If no PTP was set, the second call applies more structured urgency:

• States credit bureau reporting timeline: "Accounts overdue for 30+ days are reported to CIBIL. Your credit score is currently being affected."
• Offers immediate resolution options with clear value proposition
• Sets firm deadline: "If we don't receive payment or a commitment by [date], this will be escalated to our dedicated collections team."

The Bucket-3 AI Collections Playbook (DPD 61–90)

Objective

Prevent NPA formation (which typically triggers at DPD 90 for retail loans). Bucket-3 is the last point at which AI can play a primary role — beyond DPD 90, human collections and legal processes typically take over.

Call Architecture

Bucket-3 calls require the AI to balance urgency with compliance. The key parameters:

1. Clear statement of consequences: "Your account will be classified as Non-Performing Asset on [date] — [X] days from now. This will have significant impact on your credit score and may restrict your ability to take any new loans."

1. Structured settlement offer: "To prevent NPA classification, we need at minimum ₹X paid by [date]. If you cannot pay the full overdue, we may have partial payment options — shall I explain?"

1. Full settlement negotiation trigger: For accounts where the lender has a defined settlement policy, the AI presents the settlement offer:

• "Given the situation, we have a settlement option where you pay ₹X as full and final settlement of the outstanding of ₹Y. This would close the account and we would update the credit bureau accordingly."
• Documents the offer; all settlement authority remains with human credit officers

1. SARFAESI pre-notice language (secured loans): For Bucket-3 secured loan accounts, the AI may need to communicate that the lender is preparing a notice under SARFAESI. The exact language must be approved by the legal team and must be accurate, not threatening. "We are required by law to send a formal notice regarding your property/vehicle before proceeding with recovery. You will receive this notice at your registered address within [X] days."

1. Escalation to field collections: For accounts where phone communication has consistently failed, AI escalates to field collections. The AI makes one final call to inform the borrower that a field agent will visit.

Priority Segmentation in Bucket-2 and Bucket-3

Not all accounts deserve equal AI effort. A smart Bucket-2/3 AI programme segments by:

Value: Higher outstanding amount = higher priority, faster escalation to human

Product type: Secured loans get more effort and more empathy; unsecured high-interest loans have different recovery economics

Vintage: A borrower who has been good for 3 years and just had a rough patch is different from a first-year borrower showing early stress signs

Bureau score trajectory: A borrower whose CIBIL score has been dropping for 3 months is more stressed than one with a sudden single miss

Response behaviour: Borrowers who have been answering calls but not paying are more likely strategic defaulters; borrowers who suddenly stop answering may have had a life event

AI models can integrate all of these signals to create a priority score that determines call frequency, escalation timing, and settlement authority.

Promise-to-Pay Management and Follow-Through

One of the most critical AI capabilities in Bucket-2/3 collections is PTP management:

• AI captures PTP date and amount from every call
• PTP is stored in the collections system with borrower identity, date, amount, and channel
• AI schedules automatic follow-up on PTP date
• If payment received: AI confirms and sends receipt
• If payment not received by PTP date: AI calls on day 1 and day 3 post-PTP-miss, then escalates to human
• PTP fulfilment rate is tracked as a KPI (industry benchmark: 55–65%; well-designed AI programmes achieve 65–75%)

Broken PTPs are as important as no-PTPs — a borrower who consistently promises and fails to pay is either in genuine distress (needs restructuring) or a strategic defaulter (needs escalation).

Hardship Detection and Responsible Escalation

The most important compliance and ethical requirement in Bucket-2/3 collections is detecting genuine hardship and not pursuing collections pressure against vulnerable borrowers.

AI must detect hardship signals:

Hardship detection is both a legal requirement (RBI Fair Practices Code) and an ethical obligation. AI that ignores hardship signals and continues aggressive collections is a regulatory risk for the lender.

RBI Compliance Architecture for Bucket-2/3 Collections

Lenders should conduct quarterly audits of AI call recordings against these requirements. Non-compliance in even a small percentage of calls is a regulatory risk.

The Human-AI Handoff Protocol

The Bucket-2/3 AI is not a standalone system — it is the front layer that prepares high-quality, pre-qualified cases for human collectors. The handoff protocol is critical:

What AI passes to human agent:

• Call history (dates, dispositions, audio recordings)
• Reason stated by borrower (categorised: cash flow, distress, dispute, avoidance, no contact)
• PTP history and fulfilment record
• Hardship flags
• Recommended action (restructuring, settlement, field visit, legal notice)

What human agent should NOT do:

• Start from zero — the agent should open the conversation with "I can see you spoke with our colleague on [date] and mentioned [X]. Has the situation changed?"
• Repeat the same script the AI has already delivered — this is frustrating for borrowers and suggests the lender doesn't remember them

The warm handoff from AI to human, with full context, is what differentiates a professional collections operation from an assembly line.

Performance Metrics for Bucket-2/3 AI Collections

FAQ

Q1: Should AI be making Bucket-3 collections calls, or is this too late for AI? AI is highly effective even in Bucket-3 — the combination of consistent coverage, documented PTP capture, and compliance-safe language makes AI valuable through DPD 90. Human agents handle escalations, but AI maintains the high-frequency touchpoints that keep the account from silently sliding into NPA.

Q2: What language should AI use when communicating SARFAESI implications? SARFAESI-related language must be approved by the lender's legal team. The AI reads exact, pre-approved language — it does not paraphrase legal notices. The tone should be informational ("you will receive a notice") rather than threatening ("we will take your property").

Q3: How does AI manage a borrower who has borrowed from multiple lenders and is prioritising payments? AI cannot know the borrower's full debt profile, but it can help the lender's account remain top-of-mind and top-of-priority. Clear, respectful communication about consequences, combined with practical payment options, improves the chance that the borrower prioritises this lender's EMI.

Q4: Is it appropriate to use settlement offers in Bucket-3? Yes, where the lender's credit policy permits. AI can present settlement offers (haircut from outstanding balance) and document the borrower's response. Actual settlement approval requires human credit authority. AI accelerates the qualification and documentation process.

Q5: How does AI handle co-borrowers and guarantors in Bucket-2/3? The lender's legal team and credit policy determine when co-borrowers and guarantors are contacted. AI can make separate calls to co-borrowers (with independent authentication), with appropriate language that respects their separate legal status. Guarantor calls should only be made after defined triggers are met per RBI guidelines.

Q6: What is the maximum call frequency for Bucket-2 and Bucket-3 accounts? RBI's Fair Practices Code does not specify exact frequency limits, but general industry practice is 1–2 calls per week for Bucket-2 and 3–4 calls per week for Bucket-3, spread across different times of day. The AI should cap total call attempts and move to SMS/WhatsApp follow-up if calls go unanswered after 3 attempts in a week.

Conclusion

Bucket-2 and Bucket-3 collections represent the most consequential — and most complex — segment of the collections lifecycle. AI that handles this segment effectively combines empathy, urgency, regulatory compliance, and intelligent escalation in ways that human teams struggle to maintain consistently at scale.

The lenders that invest in sophisticated Bucket-2/3 AI will see measurable NPA reduction, lower collections cost, better compliance posture, and — critically — a collections process that identifies and protects genuinely distressed borrowers while firmly pursuing those who can pay.

Explore YuVerse's AI-powered Bucket-2 and Bucket-3 collections solutions.