The Company
PremiumFin Capital is a mid-size Non-Banking Financial Company (NBFC) headquartered in Mumbai, with operations across Maharashtra, Gujarat, Rajasthan, and Madhya Pradesh. The company specializes in personal loans, two-wheeler loans, and small business loans, with a total active loan book of approximately INR 2,800 crore across 4.2 lakh active accounts.
PremiumFin serves primarily the "new to credit" segment -- first-time borrowers who may not have extensive credit histories. This customer segment tends to have higher delinquency rates but also represents a large, underserved market opportunity.
The Challenge
Rising Delinquency, Rising Costs
PremiumFin's delinquency rates had been climbing steadily:
- 1-30 days past due (DPD): 12.4% of active accounts
- 31-60 DPD: 6.8%
- 61-90 DPD: 3.2%
- 90+ DPD (NPA): 2.1%
The total amount in delinquent accounts was approximately INR 168 crore, with the 1-30 DPD bucket representing the largest opportunity for early intervention.
The Collection Operation
PremiumFin operated an in-house collection team of 85 agents plus relationships with 3 external collection agencies. The operation faced several structural problems:
High cost per collection: The fully loaded cost of the in-house collection operation was INR 1.42 crore per month. External agencies charged 8-12% of the collected amount. Combined, the cost per successful collection was INR 380 for early-stage (1-30 DPD) accounts and INR 920 for mid-stage (31-90 DPD) accounts.
Agent productivity ceiling: Each human agent could make 80-100 calls per day, of which only 35-45 resulted in right party contact (RPC). An agent working a full 8-hour shift could have meaningful collection conversations with approximately 20-25 borrowers per day.
Compliance risk: Despite regular training, compliance violations were a persistent concern. RBI audits had flagged three incidents in the previous year involving agents using inappropriate language or calling outside permitted hours. Each incident required remediation, additional training, and management attention.
Multilingual limitations: PremiumFin's borrowers spoke Hindi, Marathi, Gujarati, and Rajasthani (with Hindi/English mixing common). The agent pool was predominantly Hindi-speaking, which created friction with Marathi and Gujarati borrowers.
Inconsistent follow-up: With 52,000 accounts in the 1-30 DPD bucket, each agent was assigned 600+ accounts. Systematic, multi-touch follow-up for every account was impossible. Agents prioritized larger balances, leaving smaller accounts uncontacted until they rolled into later delinquency buckets.
The Strategic Problem
PremiumFin's leadership recognized that their collection operation was creating a negative cycle:
- Insufficient agent capacity meant many early-stage delinquent accounts went uncontacted
- Uncontacted accounts rolled to later delinquency stages where collection was harder and more expensive
- Higher delinquency rates required more provisioning, reducing capital available for new lending
- More agents were hired, but attrition (55% annually) meant the team was constantly in training mode
Breaking this cycle required either dramatically increasing capacity (unaffordable) or finding a more efficient way to contact and collect from early-stage delinquent accounts.
The Solution
PremiumFin deployed AI voice agents for early-stage collection (1-30 DPD and 31-60 DPD accounts), keeping human agents focused on mid-to-late stage collection where negotiation skill and empathy matter most.
Deployment Scope
- Phase 1: 1-30 DPD accounts (52,000 accounts, personal loans only)
- Phase 2: 1-30 DPD accounts (all loan products)
- Phase 3: 31-60 DPD accounts (all loan products)
- Human agents continued to handle 61+ DPD accounts exclusively
Compliance-First Design
Given the regulatory environment for collections in India, the solution was designed with compliance as the primary requirement, not an afterthought.
Built-in compliance controls:
- TRAI calling hours: Hard enforcement of 9 AM - 9 PM calling window based on borrower's registered location
- NDNC scrubbing: Automated scrubbing against the national DNC registry before every campaign
- Frequency limits: Maximum 2 call attempts per day, 5 per week per account
- RBI-compliant disclosure: Every call opens with institution identification, purpose, and recording disclosure
- Language controls: No threatening, abusive, or coercive language in any script variant -- verified through compliance review and ongoing monitoring
- Third-party protection: If someone other than the borrower answers, the AI discloses only that it is calling from PremiumFin and asks to speak with the named borrower. No debt details are shared with third parties.
- Dispute handling: If a borrower states a dispute, the AI logs the dispute, pauses further collection calls, and triggers the dispute resolution workflow
Read the full guide on AI voice agent compliance for debt collection in India.
Conversation Design
The AI voice agent used a graduated intensity approach based on DPD stage:
1-7 DPD (Gentle Reminder):
"Hello [Name], this is [Agent Name] calling from PremiumFin Capital regarding your personal loan account ending in [XXXX]. This call may be recorded. Your EMI of INR [amount] was due on [date]. This is a friendly reminder. Would you like me to send you a UPI payment link so you can complete the payment at your convenience?"
8-15 DPD (Informational):
"Hello [Name], this is [Agent Name] from PremiumFin Capital. I am calling about your personal loan EMI of INR [amount] that was due on [date] and is now [X] days overdue. I want to make sure there are no issues with your payment. Would you like to discuss the available payment options?"
16-30 DPD (Concerned, Firm):
"Hello [Name], this is [Agent Name] from PremiumFin Capital. Your EMI payment of INR [amount] is now [X] days past due. It is important to resolve this soon to protect your credit score and avoid additional charges. I have several payment options that can help. Can we discuss what works best for your situation?"
31-60 DPD (Escalated, Empathetic):
"Hello [Name], this is [Agent Name] from PremiumFin Capital. I am reaching out about your overdue balance of INR [total outstanding] on your loan account. I understand that circumstances can be challenging, and I am here to find a solution that works for you. PremiumFin offers restructuring options including extended payment plans. Would you like to discuss these?"
Each script variant was reviewed by PremiumFin's legal and compliance team before deployment.
Technical Integration
- Loan Management System (LMS): Real-time access to borrower data, loan details, payment history, and outstanding amounts
- Payment gateway: Generation and delivery of UPI and net banking payment links via SMS during the call
- CIBIL/credit bureau reporting: Integration to verify and communicate credit score implications accurately
- Dispute management system: Automated logging of disputes with workflow triggers
- Campaign management: Automated bucketing by DPD stage, prioritization by balance amount, and scheduling by optimal contact time
Results
6-Month Performance Data
| Metric | Before (Human Only) | After (AI + Human) | Change |
|---|---|---|---|
| 1-30 DPD Metrics | |||
| Accounts contacted per month | 18,200 (35% of pool) | 48,700 (94% of pool) | +168% |
| Right party contact rate | 42% | 48% | +14% |
| Promise to pay (PTP) rate | 34% | 38% | +12% |
| PTP kept rate | 58% | 64% | +10% |
| Roll rate (1-30 to 31-60 DPD) | 32% | 21% | -34% |
| Cost per successful collection | INR 380 | INR 185 | -51% |
| 31-60 DPD Metrics | |||
| Accounts contacted per month | 8,400 (44% of pool) | 16,800 (88% of pool) | +100% |
| Roll rate (31-60 to 61-90 DPD) | 28% | 19% | -32% |
| Cost per successful collection | INR 920 | INR 440 | -52% |
| Compliance Metrics | |||
| Calling hour violations | 3-5 per month | 0 | -100% |
| Complaint rate | 0.18% of contacts | 0.04% of contacts | -78% |
| Audit findings | 3 in prior year | 0 in 6 months | -100% |
Financial Impact
| Category | Monthly Impact |
|---|---|
| Collection Improvement | |
| Additional collections from higher contact rate (1-30 DPD) | INR 3.8 crore |
| Additional collections from higher contact rate (31-60 DPD) | INR 1.2 crore |
| Reduced roll-forward (lower provisioning requirement) | INR 2.1 crore (capital benefit) |
| Cost Reduction | |
| Reduced in-house agent requirement (85 to 48 agents) | INR 62 lakh saved |
| Reduced external agency fees | INR 28 lakh saved |
| Platform Cost | |
| AI voice agent platform | INR 18 lakh/month |
| Net Monthly Benefit | INR 7.72 crore |
The ROI exceeded projections by 3x, primarily driven by the higher contact rate on early-stage accounts. Contacting 94% of 1-30 DPD accounts (up from 35%) significantly reduced roll rates, which had a compounding positive effect on the overall delinquency book.
Organizational Impact
Human agent redeployment: The in-house team was reduced from 85 to 48 agents through natural attrition. The remaining agents were redeployed to handle 61+ DPD accounts where negotiation, empathy, and payment restructuring require human judgment. These agents received additional training in negotiation and were given higher per-account compensation for successful resolutions.
Management bandwidth: Collection managers previously spent significant time on compliance monitoring, training, and managing agent performance for routine calls. With AI handling early-stage collection, management focused on strategy, agent skill development, and relationship management with external agencies.
External agency optimization: PremiumFin renegotiated its external agency contracts, reducing the number of agencies from 3 to 1 and focusing the remaining agency exclusively on NPA recovery. The AI voice agent handled all pre-NPA work.
Key Implementation Decisions
Decision 1: AI Identifies Itself
PremiumFin chose to have the AI identify itself as "calling from PremiumFin Capital" without explicitly stating it is an AI. The legal team determined this was compliant with current regulations. However, a disclosure framework was prepared in anticipation of future AI transparency requirements so it could be activated immediately if regulations change.
Decision 2: UPI Payment Links During the Call
The ability to send a UPI payment link via SMS during the call was a critical conversion driver. The flow: AI mentions the payment link, sends it, and asks the borrower to check their SMS. In 23% of successful collections, the borrower made the payment during or immediately after the call, directly attributable to the convenience of the UPI link.
Decision 3: Graduated Intensity by DPD Stage
Rather than a single collection script, the team designed four distinct conversation tones based on DPD stage. This was important because:
- A borrower who is 3 days past due needs a gentle nudge, not a stern warning
- A borrower who is 45 days past due needs a more urgent conversation with payment plan options
- Using the wrong tone at the wrong stage damages the relationship and reduces future collection probability
Decision 4: Keep Human Agents for 61+ DPD
The team evaluated using AI for later-stage collection but decided that accounts past 61 DPD typically involve genuine financial difficulty, disputes, or willingness-to-pay issues that require human judgment. The AI was more effective at high-volume, early-stage intervention where the primary goal is to remind and facilitate payment.
Lessons Learned
1. Contact Rate Is the Biggest Lever
The single most impactful improvement was not the AI's conversation quality or the payment link feature -- it was simply contacting more borrowers. Moving from 35% to 94% contact coverage on 1-30 DPD accounts was the primary driver of the 34% reduction in roll rates. Many borrowers in this bucket simply forget to pay and need a reminder.
2. Timing Matters for Collection Calls
A/B testing revealed optimal calling times:
- Salaried borrowers: Best contact and payment rates on salary day +1 and +2 (typically 1st-3rd of the month)
- Self-employed borrowers: Midweek (Tuesday-Thursday), 11 AM - 1 PM
- Young borrowers (under 30): Evening calls (5-8 PM) had 40% higher connection rates than morning calls
The AI campaign scheduler was configured to call each borrower segment at the optimal time.
3. Compliance as a Competitive Advantage
Zero compliance violations in 6 months became a selling point with banking partners. PremiumFin's collection compliance record improved its standing in co-lending arrangements and partnership discussions with larger banks.
4. Small Balances Add Up
Previously, accounts with EMIs below INR 2,000 were deprioritized because the collection cost per agent interaction was not justified. With AI handling these calls at INR 8-12 per attempt, even INR 500 EMI collections were economically viable. The aggregate recovery from previously neglected small-balance accounts was INR 48 lakh per month.
Key Takeaways for Financial Services
-
The primary ROI driver is contact coverage, not conversation quality. Getting from 35% to 94% contact rates on early-stage delinquency delivers more value than any improvement to the conversation script.
-
AI works best for early-stage collection where the primary task is reminding and facilitating payment. Human agents should be preserved for later-stage work requiring negotiation and empathy.
-
Compliance is a feature, not a constraint. AI voice agents deliver perfect compliance on every call, eliminating the regulatory risk that plagues human-only collection operations.
-
UPI integration is essential for Indian collections. Sending a payment link during the call captures intent at the moment of maximum willingness to pay.
-
Roll rate reduction has compounding effects. Every account that is resolved at 1-30 DPD instead of rolling to 31-60 DPD reduces future collection costs and provisioning requirements.
Model the impact for your lending portfolio using the ROI Calculator. For a detailed guide on regulatory compliance for AI collection calls in India, read the TRAI/RBI compliance guide.