Collections & Recovery Workforce for Lenders

For most NBFCs, collections is where portfolio quality is either defended or lost. If collections is slow, reactive, and undifferentiated, NPA rises regardless of sourcing quality.

What are the core failures of manual collections?

Traditional models are reactive by design, undifferentiated across borrower segments, and operationally manual at scale. These three weaknesses compound as the overdue book grows.

Collections Failure Mode	Cost to NBFC	AI Solution
Reactive intervention after default	Lost prevention window, higher NPA formation	Early Warning System (EWS)
Undifferentiated treatment by DPD	Wasted effort, under-focus on high-risk accounts	Predictive intent and ability scoring
Manual coverage at scale	Rising cost-per-recovery and burnout	Voice bots and workflow automation
Unoptimised field operations	Low visit productivity, high cost	AI route planning and prioritisation
Ad-hoc settlement decisions	Inconsistent outcomes and below-floor recoveries	Policy-banded settlement engine

AI-augmented collections consistently improves pre-default recovery economics and reduces avoidable cost in early buckets.

An early warning system (EWS) continuously tracks pre-default behavior and flags at-risk accounts 30-60 days before missed EMI.

EWS shifts collections from reactive cleanup to preventive risk action, which is usually the highest-ROI layer in collections tech.

Six high-signal indicators

• Account balance decline: liquidity stress visible before EMI impact.
• New bureau enquiries: potential refinancing or leverage stress.
• Contact behavior changes: avoidance patterns in calls and message response.
• NACH pre-notification failure: balance insufficiency before debit presentation.
• UPI activity drop: business slowdown signal in self-employed segments.
• GST filing gaps: potential business distress for SME borrowers.

LendingIQ: builds Early Warning System that combines bureau, AA, NACH, and contact behavior into a daily-updated risk score with automated workflow triggers completely customized for your Lending Organization.

Predictive scoring assigns intent-to-pay, ability-to-pay, and self-cure probability for each account. This allows effort allocation by expected outcome, not DPD alone.

Segment	Intent	Ability	Recommended Action
Green	High	High	Light nudge, likely self-cure
Amber	High	Low	Restructure discussion
Red	Low	High	Firm escalation track
Black	Low	Low	Settlement or provisioning pathway

Operational principles

1. Update intent score after contact using transcript and sentiment outcomes.
2. Set contact frequency by segment to avoid over-contacting high self-cure accounts.
3. Use vintage cohorts to identify deteriorating disbursement months early.
4. Integrate live ability signals (where consented AA data is available).

Voice bots are legally usable in collections in India when RBI Fair Practice Code controls are embedded in system logic, not left to manual discretion.

Non-negotiable compliance controls

• Calls only between 8 AM and 7 PM local time.
• Bot disclosure within first 5 seconds of call.
• Immediate human transfer option at any point.
• DND scrubbing before campaign execution.
• Complete retrievable call logs with outcomes.
• Grievance redressal details available on request.

Three-branch call architecture

1. 1
   Engagement branch: borrower acknowledges; send payment link instantly.
2. 2
   Already-paid branch: verify against records; route unresolved to reconciliation.
3. 3
   Cannot-pay branch: empathy script plus restructure or human callback track.

Performance should be measured by resolution rate and payment commitment quality, not call completion count.

Collections strategy should change by DPD stage because borrower psychology, options, and cost-of-recovery materially shift across buckets.

Pre-DPD prevention

Run reminders and EWS-triggered proactive outreach before EMI miss to avoid avoidable DPD entry.

DPD 1-30 soft intervention

Focus on payment facilitation, hardship discovery, and in-bucket resolution to prevent roll to 60.

DPD 31-60 active recovery

Increase intervention intensity, schedule field visits where justified, and prepare legal escalation for unresponsive high-risk accounts.

DPD 90+ resolution

Move to formal NPA workflows, policy-led settlements, legal recovery paths, and recovery agency assignment where needed.

Classification and provisioning governance should align with RBI asset classification directions.

Field collections remains critical for selected secured and high-ticket accounts, but unoptimised field models are expensive and low-yield.

Four optimisation principles

1. 1
   Prioritise by expected recovery value. Filter out visits where cost likely exceeds outcome.
2. 2
   Geo-cluster routes. Increase productive visits per agent through travel-time minimisation.
3. 3
   Capture outcomes in real time. Enable same-day replanning and substitution.
4. 4
   Measure conversion, not volume. Payment outcome per visit is the primary KPI.

An AI settlement engine computes policy-compliant offers that maximise expected recovery while preserving consistency and governance.

Four-step architecture

1. 1
   Define settlement bands by product and DPD. Hard policy floors, not discretionary guidance.
2. 2
   Calibrate offer inside band using ability signals. Position offers by current repayment capacity and risk context.
3. 3
   Auto-generate settlement documentation on acceptance. Eliminate drafting delay and term inconsistency.
4. 4
   Track conversion and refine quarterly. Improve offer precision with outcome feedback loops.

AI-first collections teams separate volume execution from judgment and accountability layers.

Recommended structure

• Tier 1 - AI automation: EWS monitoring, pre-DPD nudges, early bucket outreach, payment link delivery, score updates.
• Tier 2 - Human specialists: high-value escalations, restructure negotiation, field supervision.
• Tier 3 - Legal and recovery: SARFAESI and legal resolution workflows, agency orchestration.
• Tier 4 - Compliance oversight: bot audit, DND checks, settlement policy adherence, escalation quality review.

AI handles consistency and scale; humans handle nuance, legal accountability, and relationship judgment.

LendingIQ builds Collections Workforce that does pre-DPD risk detection to NPA resolution, with RBI-aligned controls and Account Aggregator integration support completely customized for your Lending Organization.

Related reading

• Risk Workforce for Lenders: AI Credit, Fraud & Portfolio Management
• NBFC Compliance & Regulatory Readiness: Collections and Recovery Requirements
• Account Aggregator Integration Guide: Using Consent Data in Collections