Debt recovery in the 31–90 DPD window requires more than reminders — it requires structured negotiation: understanding why the borrower hasn't paid, identifying which resolution path fits their situation, and securing a firm commitment with consequences clearly understood. Human collectors do this well at low scale; AI does it consistently at high scale.

Kallix AI debt recovery agents are configured by the lender's collections head with a negotiation decision tree: which settlement offers apply at which DPD stage, what partial payment thresholds are acceptable, when to present OTS vs EMI restructuring vs tenure extension, and when to route to a specialist human negotiator. The AI follows this tree precisely — no improvisation, no inconsistent offers, no harassment.

Key functions: (1) structured opening with identity verification and account summary; (2) reason-for-non-payment capture — a free-form exchange that classifies the borrower into hardship profiles (job loss, medical, business failure, dispute, wilful defaulter); (3) offer presentation based on the hardship profile and DPD count; (4) objection handling with scripted rebuttals drawn from the lender's approved response library; (5) PTP capture with explicit confirmation; (6) SARFAESI/legal consequence advisory for accounts approaching 90 DPD.

Kallix production data across 80+ lender customers shows AI-assisted recovery calls improve the 31–60 DPD resolution rate by 18–28% compared to human-only dialler campaigns with the same contact volume — primarily because AI achieves higher right-party contact rates and never has a 'bad call day.'

The DPD segmentation in Indian retail collections is shaped by both commercial practice and RBI regulation:

**0 DPD (pre-delinquency)**: Pre-debit reminders — Kallix soft collections module.
**1–30 DPD (SMA-0 and SMA-1)**: Soft collections — automated nudges, UPI payment links, NACH bounce follow-up.
**31–60 DPD (SMA-2)**: Hard collections — structured negotiation calls, settlement offers, PTP capture. Kallix debt recovery module.
**61–90 DPD (SMA-2 continuing)**: Pre-NPA — OTS offers, legal consequence advisory, escalated negotiation. AI with parallel human specialist involvement.
**90+ DPD (NPA/Sub-Standard)**: NPA classification by the lender. AI transitions to a support role: automated payment-link messages, settlement reminders. Primary recovery shifts to legal and field teams.

For NBFCs (as opposed to scheduled commercial banks), the SMA classification thresholds differ slightly — NBFCs classify SMA-1 at 31–60 days and SMA-2 at 61–90 days per RBI NBFC Master Directions. Kallix's configuration supports both bank and NBFC SMA frameworks.

Important boundary: AI voice agents conducting recovery calls must not misrepresent their legal authority or imply imminent legal action unless the lender has actually initiated it. The AI states consequences factually ('accounts classified as NPA at 90 days may be subject to SARFAESI proceedings') without implying the lender has already decided to pursue legal action for a specific account.

OTS negotiation is the highest-value recovery action for 60–90 DPD accounts — it converts a potential NPA write-off into a cash recovery, even at a discount. But it is also the most legally sensitive: OTS terms must be documented, sanctioned by the lender's credit committee, and executed with proper waivers signed. AI handles the identification and initial offer; humans handle the execution.

Kallix's OTS flow:
1. **Eligibility check**: the AI reads the account's OTS eligibility flag from the LMS (set by the lender's credit team). Not all accounts receive OTS offers — the flag gates access.
2. **Offer presentation**: the AI states the pre-approved OTS amount, validity period, and payment method ('full payment via RTGS/NEFT within 15 days to close the account and receive a No-Dues Certificate').
3. **Borrower response handling**:
- Acceptance: AI logs acceptance, triggers a human callback within 2 hours for documentation, sends OTS offer letter via email/WhatsApp.
- Counter-proposal (lower amount): AI states it cannot accept counter-proposals autonomously and transfers to a human negotiator with the counter-proposal logged.
- Decline: AI logs decline reason, notes DPD, and schedules a follow-up call in 7 days if DPD is below 85 (within 5 days of NPA cutoff).
4. **CIBIL outcome advisory**: The AI proactively explains that OTS will result in CIBIL status 'Settled' (not 'Closed') — a distinction that significantly impacts future credit access. This transparency is both ethical and builds trust.

For secured loans (home, vehicle), the AI also mentions the collateral release process — that upon OTS payment, the lender will release the title deed/RC within the timeline specified in the loan agreement, typically 15–30 working days.

The Settled vs Closed distinction is one of the most misunderstood aspects of loan resolution in India, and it directly affects borrower decisions. The Kallix debt recovery agent explains this proactively rather than waiting for the borrower to ask — it is one of the most effective reasons to drive full payment over discounted settlement when the borrower has partial capacity.

**'Closed' status**: Reported when the borrower repays the full outstanding principal, interest, and applicable charges per the original loan agreement or an extended restructured agreement. Credit bureaus show this as a satisfactorily closed account. No negative signal — in fact, a 'Closed' loan improves the credit mix and repayment history components of the CIBIL score.

**'Settled' status**: Reported when the lender accepts less than the total legal outstanding — an OTS, a waiver of part of the interest, or any negotiated closure below full dues. CIBIL records this as 'Settled' and it remains visible for 7 years from the settlement date. Future lenders can see this and typically: (a) reject the borrower for premium credit products; (b) charge 100–300 bps higher interest; (c) require higher co-applicant income or collateral. Recovery from a 'Settled' tag typically takes 2–3 years of clean payment history to partially offset.

For borrowers who are 60–75 DPD and have partial capacity to pay — say, they can pay 70% of outstanding now and the remaining 30% in 3 months — the AI models the option: 'If you can arrange the full Rs 2,40,000 in two tranches — Rs 1,70,000 now and Rs 70,000 in 3 months — we can close the account as 'Closed' rather than 'Settled', which is significantly better for your future loan applications.' This negotiation path, when available in the lender's policy, increases total recovery while giving the borrower a genuine incentive.

Objection handling is where most automated collection systems fail — they play a static script and cannot adapt to what the borrower actually says. Kallix trains objection response branches specifically for recovery calls, with escalation triggers for situations the AI is not permitted to handle.

**Objection 1 — 'I'll pay next month / in a few days'**:
AI response: 'I understand. To make sure we track this correctly — what specific date can you commit to, and will that be the full amount of Rs 18,500 or a partial payment?' Converts vague intent into a PTP with a date. If borrower refuses to name a date, the AI flags as 'vague promise' in the LMS — different follow-up cadence from a firm PTP.

**Objection 2 — 'I lost my job / medical emergency / business failure'**:
AI response activates hardship protocol: acknowledges situation, offers available restructuring options (EMI deferral, step-down EMI), captures nature of hardship in LMS, transfers to loan modification specialist. Call flagged as hardship — collection intensity is reduced during a 15-day review window.

**Objection 3 — 'The amount is wrong / I don't owe this'**:
AI immediately stops collection discussion: 'If you believe there is an error in your account, I'll raise a dispute flag and pause collection calls on this account. A loan statement will be sent to your registered email within 2 working days.' Collection calls suppressed for 7 working days pending dispute resolution.

**Objection 4 — 'I already paid this'**:
AI asks for UPI/NEFT reference number, places account in 4-hour payment verification hold, sends confirmation request to borrower's email. If borrower cannot provide reference, the AI explains that unconfirmed payments will be verified within 24 hours and asks for proof via WhatsApp.

**Objection 5 — 'I can't afford to pay the full amount'**:
AI transitions to partial payment path or OTS eligibility check: 'What amount can you pay today? Even a partial payment stops further late charges from accruing and keeps your account active.' If partial payment is within lender's policy threshold (configurable), AI accepts and logs it.

**Objection 6 — 'Stop calling me / I'll take legal action'**:
AI acknowledges: 'I've noted your request. I'll flag this call for our nodal officer and you will receive a written communication at your registered address within 3 working days.' Opt-down logged in LMS, supervisor notified, collection calls suppressed for 48 hours. Lender's grievance officer pathway disclosed.

SARFAESI (Securitisation and Reconstruction of Financial Assets and Enforcement of Security Interest Act, 2002) is the primary legal enforcement tool for secured loan recovery in India. It applies to scheduled commercial banks and most NBFCs for loans above Rs 1 lakh where a security interest exists. Communicating SARFAESI consequences to a borrower is both legally required (fair notice) and commercially effective — but must be done accurately and without coercion.

Kallix's SARFAESI advisory framework:

**Trigger**: The SARFAESI advisory script activates at 75+ DPD for secured loans (home loan, vehicle loan, loan against property, SME secured loan).

**What the AI says**: 'Your account is [X] days overdue. If the account is not brought current or a payment arrangement is not made within 15 days, your loan will be classified as a Non-Performing Asset (NPA). Under the SARFAESI Act 2002, [Lender Name] has the right to issue a written demand notice. If payment is not made within 60 days of that notice, the lender may initiate possession of the secured asset. I'm telling you this so we can find a resolution before that stage.'

**What the AI does NOT say**: It does not say the lender has decided to invoke SARFAESI for this specific account unless the lender has specifically flagged the account as 'SARFAESI initiated' in the LMS. Implying imminent action that has not been decided is a misrepresentation that violates RBI Fair Practices Code.

For unsecured loans (personal loan, credit card, education loan without collateral), SARFAESI does not apply. For these, the AI communicates the alternative — NPA classification affecting CIBIL, potential DRT (Debt Recovery Tribunal) proceeding for amounts above Rs 20 lakh (banks) or civil suit for smaller amounts, and the practical difficulty of future credit access.

Section 138 of the Negotiable Instruments Act (dishonoured cheque) applies when post-dated cheques are part of the loan documentation — the AI informs relevant borrowers that cheque dishonour can be a criminal offence (up to 2 years imprisonment or 2x cheque amount fine), which is a distinct and often underutilised consequence advisory.

The RBI Recovery Agent Guidelines 2008 were issued in response to documented harassment of borrowers by recovery agents — physical intimidation, calling at odd hours, contacting employers and relatives, using abusive language. These incidents gave rise to strict regulatory standards that Kallix builds into every recovery agent deployment.

**Mandatory compliance elements and Kallix implementation**:

1. **Lender identification at call start**: 'This is an automated recovery call from [Lender Name] regarding loan account ending [XXXX].' No alias, no disguised identity.

2. **No third-party disclosure**: If a third party answers, the AI provides only the lender's callback number and says 'Please ask [Borrower First Name] to call us at [number].' No account details, no DPD count, no amount disclosed.

3. **7 AM–7 PM calling window**: Enforced at telephony routing — calls physically cannot be placed outside this window. This is a hard system control, not a script instruction that could be overridden by configuration error.

4. **No abusive, threatening, or harassing language**: All recovery scripts are reviewed by a compliance officer before deployment. The AI has no ability to generate ad-hoc language — it operates from a fixed, pre-approved script tree. Any out-of-scope conversation (e.g., a borrower using abusive language) triggers an immediate: 'I'm unable to continue this conversation. Please call our customer service at [number] to discuss your account.' followed by call termination.

5. **IBA certification for field recovery**: IBA-certified recovery agents are required for field visits — Kallix does not conduct field visits. Physical recovery is outside the AI's scope entirely.

6. **Grievance mechanism**: Every call ends with an offer of the lender's grievance officer contact details if the borrower requests. This is required by the guidelines and is built into every recovery call script as a standard closing option.

Hardship classification is what separates a collections system that recovers money from one that damages the lender-borrower relationship permanently. The correct recovery path for a borrower who lost their job is completely different from the correct path for a borrower who has money but is choosing not to pay — and treating both the same way is both ineffective and potentially non-compliant.

Kallix's five hardship profiles:

**Profile 1 — Cash Flow Timing (most common, ~45% of 31–60 DPD)**: Borrower has income but faced a short-term timing mismatch — salary delayed, client payment pending, seasonal business dip. Indicators: borrower mentions a specific date or event. Recovery path: PTP capture with a specific date, UPI payment link, follow-up call 24 hours before PTP date.

**Profile 2 — Income Disruption**: Borrower has lost their primary income source — job loss, business closure, medical incapacitation. Indicators: mentions job loss, closure, hospital, income stopped. Recovery path: restructuring advisory, loan modification officer callback within 4 hours, collection intensity reduced during 15-day review. LMS flagged as hardship case.

**Profile 3 — Dispute**: Borrower believes the amount or terms are incorrect. Indicators: 'I didn't take this loan', 'the amount is wrong', 'this loan was restructured'. Recovery path: collection suspended, dispute flag raised, loan statement dispatched, human investigation initiated.

**Profile 4 — Wilful Avoidance**: Borrower acknowledges the debt and ability to pay but refuses. Indicators: no claim of hardship, deflection without specific reason, requests to stop calling without financial explanation. Recovery path: consequence advisory (SARFAESI/NPA/CIBIL), escalation to senior human collector, shortened follow-up cadence.

**Profile 5 — Severe Hardship / Distress**: Borrower mentions death of primary earner, terminal illness, natural disaster, domestic violence, or other acute crisis. Indicators: mentions of funeral, hospital, flood, police, abuse. Recovery path: immediate compassionate escalation to human collector, collection calls paused for 30 days, case flagged for lender's hardship waiver review team.

Deceased borrower cases require immediate, compassionate handling — continuing collection calls after being informed of a death is both ethically wrong and a serious regulatory violation. Kallix trains the AI to detect deceased-borrower signals and respond appropriately.

Detection signals: respondent says 'he/she passed away', 'died', 'no more', 'expired' or uses past tense consistently to refer to the borrower ('he used to bank there').

AI response on deceased identification: 'I'm sorry to hear this. Please accept our condolences. I am stopping this automated call immediately. Our team will reach out to the appropriate contact regarding the loan account in due course. Is there a number where our customer care officer can contact the family?'

LMS actions triggered: (1) all automated calls to the registered mobile suppressed permanently; (2) 'deceased — legal heir process' flag raised; (3) human recovery officer assigned; (4) lender's legal team notified if the outstanding is above the lender's threshold for legal heir recovery.

Legal heir recovery framework: In India, legal heirs are not personally liable for the deceased borrower's debts unless they were co-applicants or guarantors. However, if the deceased had a life insurance policy attached to the loan (increasingly common for home and vehicle loans), the lender can claim from the insurer. The human recovery officer's job is to: (a) verify death and obtain death certificate; (b) identify co-applicant or guarantor; (c) check for loan insurance; (d) if no co-applicant and no insurance, pursue recovery from the estate through civil channels if warranted.

Kallix does not handle the legal heir recovery process itself — it manages the trigger and handover cleanly.

Partial payment thresholds in recovery are a credit policy decision — accepting too-small a partial payment can: (a) reset the DPD clock in ways that complicate SARFAESI eligibility tracking; (b) signal to the borrower that further discounting is possible, undermining the OTS floor; (c) create accounting complexity for accrued interest. Kallix therefore enforces the lender's configured minimum partial payment as a hard threshold.

The AI's response to a below-threshold offer: 'Thank you for the offer to pay Rs 5,000 today. Our minimum partial payment for this account is Rs 8,000. Would you like to proceed with Rs 8,000? Alternatively, if you're finding the full outstanding difficult, I can share a one-time settlement option that may be more manageable.'

If the borrower insists on the lower amount ('I only have Rs 5,000, I cannot do more'), the AI does two things: (1) logs the offer and the borrower's stated maximum capacity in the LMS; (2) escalates to a human negotiator who has authority to make a policy exception with documented justification.

The key principle: the AI never rejects a borrower's payment offer with a flat 'no'. It always presents an alternative path — the minimum partial payment, the OTS, or the human escalation — so the conversation ends with a clear next step rather than a dead end.

For credit card recovery specifically, minimum payment (as distinct from full outstanding) has different implications — paying the minimum due stops the DPD counter but does not resolve the recovery situation. The AI distinguishes clearly between 'minimum payment to bring the account current' and 'minimum partial payment toward outstanding settlement' to avoid confusion.

Multi-debt borrowers are increasingly common in India, particularly in the 31–90 DPD bucket — a borrower under cash flow stress often stops paying multiple lenders simultaneously. The AI must handle this carefully: it has authorisation only to discuss the calling lender's accounts, and cannot make representations about or negotiate on behalf of other lenders.

Within the same lender's portfolio: If a borrower has a personal loan (45 DPD), a credit card (62 DPD), and a vehicle loan (18 DPD), Kallix's LMS query at call start retrieves all accounts. The recovery call covers all three in sequence: 'I'm calling regarding three accounts with [Lender Name]. Let me walk through each briefly.' This reduces the total call volume the borrower receives from the lender and provides a consolidated view of their obligation.

For combined outstanding, the AI can present a consolidated settlement path if the lender's credit policy permits cross-product OTS — for example, settling all three accounts with a single combined payment.

When the borrower raises other-lender debts: 'I'm already getting calls from two other banks — I can't pay everyone.' AI response: 'I understand that's a difficult situation. I can only discuss your accounts with [Lender Name] — I'm not in a position to negotiate on behalf of other lenders. For your [Lender] account specifically, here's what we can do to help...'

Debt consolidation referral: If the lender offers a debt consolidation product (a single loan to pay off multiple creditors), and the borrower expresses interest, the AI can note the interest and transfer to a product sales officer — clearly marked as a product discussion, not a collection call.

Once an account crosses 90 DPD and is classified as NPA (Sub-Standard Asset under RBI Income Recognition and Asset Classification norms), the regulatory and legal complexity increases significantly. AI voice outbound calls at this stage are not prohibited, but the primary recovery effort typically involves legal notices, field visits, and senior human negotiators — contexts where AI plays a support role rather than the lead.

Kallix NPA support functions:

**Inbound settlement line**: Borrowers who receive a SARFAESI Section 13(2) notice often want to call and negotiate — but human agents are busy, and these calls go unanswered. Kallix provides a dedicated inbound IVR that NPA borrowers can call to: (a) express settlement intent — triggers 4-hour human callback; (b) hear current outstanding with interest and charges (authenticated via MPIN or date of birth); (c) request a loan statement for their records; (d) request time extension before legal action.

**Settlement offer reminder via WhatsApp/SMS**: For NPA accounts where the lender has approved an OTS offer, Kallix sends weekly automated settlement offer reminders via WhatsApp — with the offer amount, validity date, and payment link. This maintains contact without the harassment risk of daily voice calls.

**Write-off vs settlement distinction**: For accounts approaching technical write-off (typically 18–24 months NPA), the AI is configured with a 'final settlement offer' script — a last outreach before the account moves to the write-off ledger, after which recovery becomes significantly harder. Borrowers may not be aware that a write-off does not extinguish the debt — it only changes how the lender accounts for it internally. The AI explains this factually.

**Auction pre-notification (secured loans)**: For SARFAESI-eligible secured loans approaching possession and auction, RBI guidelines require the lender to notify the borrower before auction. Kallix can automate these pre-auction notifications, which must be sent in writing (and which Kallix dispatches via registered WhatsApp message + email).

The integration architecture for debt recovery is more complex than for servicing calls — because the AI and human collector often work the same account within the same day, and the handover quality directly impacts recovery outcomes.

**Dialler integration models**:

- **AI-first, human-backup**: Kallix dials all 31–60 DPD accounts first. Accounts that reach 'unresolved' disposition (no PTP, no settlement interest, hardship flag) after 2 AI attempts are pushed to the human dialler queue with full context. Human agent's screen shows: last AI call date, transcript summary, hardship classification, any offers the borrower rejected.

- **Parallel dialling (blended)**: Kallix and the human dialler run simultaneously against different DPD segments. Kallix handles 31–50 DPD; human collectors handle 51–90 DPD. PTP data shared bidirectionally so both systems avoid duplicate contacts.

- **Inbound escalation**: AI call in progress; borrower requests human at any point. Warm transfer with context — the human collector hears a 10-second AI summary of the call so far ('Borrower: Rahul Sharma, 45 DPD, Rs 22,000 overdue, states job loss, requesting restructuring') before being connected to the borrower.

**CRM/LMS write-back fields** (written after every AI call):
- Call outcome (PTP / settlement interest / hardship / dispute / unresolved / no-contact)
- PTP amount and date (if captured)
- Hardship profile classification
- Settlement offer presented Y/N, borrower response
- Next action flag (human follow-up / scheduled AI call / suppressed)

Integration deployment timeline: 2–3 weeks for dialler integration (API-based); 3–4 weeks for LMS write-back mapping; 1 week UAT with live call simulation.

The ROI case for AI debt recovery is driven by three compounding effects: cost per contact reduction, contact rate improvement, and resolution rate improvement. All three move simultaneously.

**Cost per contact**:
- AI (Kallix): Rs 20–35 per call (telephony + AI + LMS write-back)
- Human recovery agent: Rs 90–130 per call (salary + overhead + dialler + supervisory cost)
- Net saving: Rs 55–95 per contact

**Contact rate improvement**:
- AI operates across 18 hours (7 AM–1 AM configurable) vs 9–10 hours for human agents
- AI makes unlimited simultaneous calls vs predictive dialler capacity constraints
- Right-party contact rate: 68–74% (AI) vs 45–55% (human predictive dialler) — because AI does not cherry-pick easy accounts and maintains consistent retry cadence

**Resolution rate improvement (31–60 DPD)**:
- 18–28% more accounts resolved per contact (PTP captured or settlement interest expressed)
- Primary driver: AI never has a bad day — script consistency means every borrower gets the same quality interaction regardless of time of day, call volume pressure, or agent fatigue

**Deployment economics (30,000 active delinquent accounts)**:
- Monthly AI calls: ~90,000 (3 per account)
- Human cost avoided: Rs 81 lakh–Rs 1.17 crore/month
- Kallix cost: Rs 27–31.5 lakh/month (including platform fee)
- Monthly net saving: Rs 49.5 lakh–Rs 85.5 lakh
- One-time setup cost (including LMS + dialler integration): Rs 10–20 lakh
- Payback period: 45–60 days

Additional recovery uplift (harder to quantify but consistently observed): 12–18% reduction in legal recovery costs because fewer accounts require SARFAESI proceedings when the 31–60 DPD window is resolved effectively.

Distress signal detection is both an ethical requirement and a regulatory one. The RBI Fair Practices Code requires that recovery agents not take advantage of borrowers in vulnerable circumstances. The AI handles this through keyword and semantic signal detection across the conversation.

**Distress signal categories**:

- **Death/bereavement**: 'My husband just died', 'we had a funeral last week', 'sole earner is no more'
- **Medical emergency**: 'I'm in the hospital', 'terminal illness', 'just had surgery', 'cancer treatment'
- **Suicidal ideation**: Any mention of self-harm, ending life, or extreme hopelessness ('I don't want to live if I have to keep dealing with this') — the AI is trained to flag these with high sensitivity, even at the risk of false positives
- **Natural disaster / force majeure**: 'My house was flooded', 'the cyclone destroyed everything', 'we lost everything in the fire'
- **Domestic violence**: 'I can't talk, my husband doesn't know I took this loan', 'I'm afraid of what will happen'

**AI response on distress detection**:
'I want to pause our conversation for a moment. It sounds like you're going through something very difficult. Please know that our team can work with you on your loan — there's no pressure right now. If you need support, you can reach iCall at 9152987821. Our customer care at [number] can also help. I'm going to pause calls on this account while we get the right support in place.'

**LMS actions**: All calls to the account suppressed for 30 days. 'Sensitive case — distress flagged' notation added. Lender's sensitive case team (or social work team, where the lender has one) notified within 2 hours.

Suicidal ideation is treated as the highest-priority flag — the Vandrevala Foundation (1860-2662-345, 24/7) and iCall helpline details are provided immediately, and the account is escalated to a human within 15 minutes.

Debt recovery deployment is more complex than servicing automation because the stakes per interaction are higher — a misconfigured script or wrong OTS offer could result in a legally binding commitment the lender cannot honour, a compliance violation, or borrower harm. Kallix therefore runs a more thorough UAT and pilot phase for recovery than for servicing.

**Phase 1 — Integration (weeks 1–2)**:
- LMS API integration: DPD data, OTS eligibility flags, hardship history, existing PTP records
- Dialler integration: AI call outcome push to human dialler queue; warm transfer protocol tested
- NACH return file and bounce history mapped for 31+ DPD context
- Payment gateway for recovery: UPI collect, NEFT details, OTS payment link generation

**Phase 2 — Configuration & Compliance Build (weeks 2–4)**:
- OTS matrix uploaded: eligibility criteria by DPD, product type, outstanding range, prior history
- Hardship protocol rules configured: which hardship profiles pause collections, for how long
- Consequence advisory scripts reviewed by lender's legal/compliance team
- TRAI TCCCPR transactional classification headers registered for debt recovery category
- Distress signal dictionary calibrated to lender's specific borrower profile

**Phase 3 — UAT (week 5)**:
- 200+ test scenarios: DPD variants, objection types, hardship profiles, OTS acceptance/counter/decline, SARFAESI advisory, deceased/distress signal, language switch, escalation paths
- Legal/compliance team validates consequence advisory language and SARFAESI wording
- Lender QA validates LMS write-back accuracy for all 15+ disposition codes

**Phase 4 — Controlled Pilot (weeks 6–7)**:
- 3,000–5,000 account cohort in 31–60 DPD range
- Weekly reporting: contact rate, PTP rate, hardship classification distribution, settlement interest rate
- Script refinement based on pilot call quality review (random 5% sample listened to by compliance officer)

**Phase 5 — Full Deployment (weeks 8–9)**:
- Full 31–90 DPD portfolio segment activated
- Human dialler queue configured for AI-escalated accounts
- Live dashboard: daily recovery metrics, DPD roll rates, PTP kept rates, OTS conversion

The AI vs human question in debt recovery is a sequencing and segmentation decision, not an either/or replacement. Kallix is designed to maximise the ROI of both — AI handles volume and consistency; humans handle complexity and judgment.

**Where AI outperforms humans in recovery**:
- Volume: AI can make 10,000 calls simultaneously; a human team of 100 agents handles 800–1,000 calls/day
- Consistency: every call follows the exact script, every compliance requirement met on every call
- Contact rate: AI maintains persistent retry cadence without fatigue or cherry-picking
- First-call PTP capture: for straightforward cash-flow-timing cases, AI captures PTPs at similar rates to human agents but at 70–75% lower cost
- Availability: AI runs 7 AM–7 PM without shift changes or absenteeism

**Where humans outperform AI in recovery**:
- Complex negotiation: multi-lender settlement, structured OTS counter-negotiation, business loan restructuring
- Reading emotional tone: a skilled human collector can detect nuance in hesitation, genuine distress vs stalling, and willingness signals that AI cannot reliably assess
- Legal complexity: when a borrower raises legal counter-claims, SARFAESI challenges, or court stay orders, human legal specialists are required
- Relationship-based recovery: for HNI or business banking customers where the lender relationship manager is involved in the recovery conversation

**Recommended model for Indian retail lenders**:
- AI first-contact on all 31–60 DPD: captures PTP, classifies hardship, presents OTS eligibility check
- Human follow-up on AI-escalated accounts (unresolved after 2 AI calls): takes context from AI, handles negotiation
- Human-led on all 61+ DPD: AI provides payment-link support and inbound settlement line alongside human leads

Identity verification in recovery calls is more sensitive than in servicing calls — the information being discussed (overdue amounts, DPD count, SARFAESI eligibility) is more damaging if disclosed to a third party. Kallix enforces a two-factor check before any account detail is mentioned.

Verification flow: (1) 'Please confirm the last 4 digits of your loan account number.' DTMF or voice accepted. (2) 'Please confirm your date of birth in DD MM YYYY format.' Matched against LMS record in real time.

If a third party answers and identifies themselves: the AI provides only the lender's callback number — 'Please ask [First Name] to call us at [number].' No mention of overdue status, loan type, or amount. No implication that there is a problem. This design prevents the scenario where a borrower's family member inadvertently learns of a financial difficulty.

For accounts at 75+ DPD where SARFAESI proceedings may have been initiated, an optional third-factor OTP can be enabled — the LMS sends an OTP to the registered mobile, which the borrower reads back. This additional verification step ensures that any SARFAESI-related communication reaches the actual borrower, not a household member, and satisfies RBI's requirement for direct borrower contact before enforcement action.

Verification failure disposition: after 2 failed attempts, the call ends. Three consecutive call-level verification failures on the same day auto-suppress further calls for 24 hours and flag the account for human review — a pattern of failed verifications can indicate the number has changed or is being answered by a family member consistently.

Post-dated cheques remain a common loan security instrument in India for personal loans, vehicle loans, and MSME loans — particularly from borrowers with lower digital literacy or those offered PDC-based EMI schemes by lenders without standing instruction mandates.

When a PDC bounces, the lender has two distinct recovery paths: (1) civil recovery under the loan agreement, and (2) criminal prosecution under Section 138 of the Negotiable Instruments Act 1881. Section 138 is frequently used by lenders as a recovery lever — a criminal complaint adds urgency that a civil demand notice does not.

Kallix's PDC recovery advisory:
- The AI states factually: 'Your post-dated cheque of Rs [amount] dated [date] has been returned unpaid. Under Section 138 of the Negotiable Instruments Act, dishonour of a cheque issued as security for a debt is a criminal offence. [Lender Name] has the right to file a complaint within 30 days of notice. We'd like to resolve this before that step is taken.'
- The AI does not say the lender has decided to file a complaint unless the LMS flags 'Section 138 complaint filed.' It states the legal right factually.
- A demand notice period of 15 days (from the date of notice) is required before filing — if the borrower pays within this window, the complaint cannot proceed. The AI advises the borrower of this payment window explicitly.
- PDC cases are escalated to a human legal officer at the 61+ DPD stage, as the decision to file Section 138 requires legal team authorisation.

For ECS/NACH mandate-backed loans, the Section 138 provision does not apply (a mandate instruction differs from a cheque) — the AI distinguishes clearly and does not misapply the criminal law advisory.

The Insolvency and Bankruptcy Code 2016 (IBC) fundamentally changes the recovery landscape for a defaulting corporate or individual borrower. Once an IBC petition is admitted by the National Company Law Tribunal (NCLT) for a corporate or Debt Recovery Tribunal (DRT) for an individual, Section 14 of IBC imposes an automatic moratorium — all debt recovery actions, including calls, legal proceedings, and security enforcement, are prohibited during the Corporate Insolvency Resolution Process (CIRP).

Kallix's IBC handling:
1. **LMS flag**: If the lender's legal team flags an account as 'IBC petition admitted' in the LMS, all Kallix calls to that account are immediately and permanently suppressed. This flag must be set by the lender — Kallix cannot independently verify NCLT admission status.
2. **Borrower disclosure during call**: If a borrower states 'I've filed for insolvency' or 'I'm under IBC proceedings,' the AI immediately responds: 'Thank you for informing us. I'll pause all automated communications immediately and note this for our legal team. You will be contacted through the appropriate legal channel.' Collection suspended, legal team notified within 1 hour.
3. **Individual insolvency (IBC Part III)**: Individual insolvency proceedings under IBC are rare but growing — the moratorium applies identically. The AI's response protocol is the same as for corporate IBC.
4. **OTS during IBC CIRP**: In some cases, lenders participate in the Committee of Creditors (CoC) process and may accept a resolution plan. This is entirely outside AI scope — a senior executive and legal team matter.

Note: mere default (even severe NPA) does not trigger IBC moratorium. The moratorium begins only upon NCLT/DRT admission of the petition, not filing.

Co-applicant and guarantor recovery calls require careful handling — they involve a different person than the primary borrower, with a potentially different relationship to the debt, different communication preferences, and full legal liability that many co-applicants do not fully understand until a recovery call arrives.

Kallix's co-applicant contact protocol:

**Trigger condition**: Configurable — default is 5 consecutive business days of no-right-party-contact with the primary borrower. The lender can adjust this to 3 or 7 days based on DPD stage.

**Script distinction**: The co-applicant call is explicitly different from the primary borrower call: 'I'm calling regarding a joint loan account you hold with [Primary Borrower Name] at [Lender Name], ending [XXXX]. As a co-applicant on this account, you are jointly and severally liable for the outstanding amount. The account is currently [X] days overdue. I'm reaching out because we have been unable to contact [Primary Borrower Name] at their registered number.'

**Legal liability advisory**: Many co-applicants (especially family members) are unaware that 'co-applicant' means full liability — not partial or secondary liability. The AI explains this clearly and without alarm: 'As a co-applicant, your credit score is affected by this account's payment status, just as the primary borrower's is. If the account reaches NPA classification, both co-applicants will have the default reflected on their bureau records.'

**Guarantor calls**: Guarantors are liable only upon borrower default and demand — a guarantor call is appropriate only after a formal demand letter has been sent to the guarantor. Kallix's LMS integration checks for 'guarantor demand letter dispatched' flag before enabling guarantor calls.

**Privacy boundary**: The AI does not disclose to the primary borrower that their co-applicant has been called, and does not disclose to the co-applicant any information about the primary borrower's financial situation beyond what is jointly relevant to the loan account.

PTP fulfilment handling is as important as PTP capture — calling a borrower who has just paid is one of the fastest ways to damage the lender-customer relationship and generate regulatory complaints. Kallix's payment confirmation architecture prevents this.

Payment confirmation flow:
1. **LMS payment match**: The LMS (or payment gateway webhook) confirms receipt of payment. For UPI payments, confirmation is near-instant (NPCI switch settlement). For NEFT/RTGS, confirmation arrives within 2–4 hours of settlement.
2. **PTP status update**: PTP record marked 'Kept' in LMS. All scheduled reminder calls (day-before, day-of) immediately cancelled.
3. **Confirmation message**: WhatsApp/SMS sent to the borrower: 'Payment of Rs [amount] received. Your loan account ending [XXXX] is now [current/partially reduced]. Thank you.' This positive reinforcement is deliberate — it acknowledges the borrower's commitment being honoured, which improves future collection cooperation.
4. **Full resolution**: If the payment brings the account to 0 DPD, the account is removed from the Kallix recovery queue entirely. No further collection calls unless a new delinquency event occurs.
5. **Partial PTP kept**: If the borrower paid Rs 8,000 of a committed Rs 12,000, the LMS records a partial keep. After a 3-day cooling period, a new PTP call is scheduled for the remaining Rs 4,000 — with a tone that acknowledges the partial payment positively before addressing the balance.

Lenders can configure the cooling period (1–7 days) based on their portfolio tolerance. A shorter cooling period increases recovery speed but risks borrower fatigue if they are genuinely unable to pay the balance immediately.

Gold loans are structurally different from other secured loans — the collateral (gold jewellery) is already in the lender's custody, which makes the recovery mechanism faster and less legally complex than SARFAESI. However, the emotional dimension is high — gold jewellery often has sentimental value that makes the threat of auction particularly effective as a resolution motivator.

Kallix gold loan recovery specifics:

**Auction timeline advisory**: 'Your gold loan of Rs [amount] is [X] days overdue. [Lender Name] policy allows auction of pledged gold if the account remains unpaid for 90 days from due date. We would like to help you retrieve your gold before that point.'

**LTV monitoring**: Gold loans are subject to LTV (Loan-to-Value) ratio monitoring — if gold prices fall and the LTV exceeds RBI's 75% cap for NBFCs or 90% for bullet repayment gold loans, the lender may require immediate part-payment or additional margin. The AI is configured to mention margin call requirements when the LMS flags an LTV breach.

**Pre-auction notice**: RBI Gold Loan Guidelines require a minimum 14-day written pre-auction notice to the borrower. Kallix automates this dispatch (via registered post WhatsApp equivalent and email) and makes a voice call on the day of notice dispatch to ensure the borrower is aware.

**Auction bidding process**: Gold is auctioned at market price minus a 10–15% discount for quick sale. The AI explains this to motivate pre-auction resolution: 'If your gold is auctioned, it will be sold at market price less a standard auction discount. Repaying the loan directly means you get your gold back at full value.'

**Post-auction surplus**: If the auction proceeds exceed the outstanding loan + interest + charges, the surplus belongs to the borrower. The AI mentions this right proactively so borrowers are not under the impression the lender profits from over-collateralised auctions.

NRI recovery calls have three distinctive challenges: time zone compliance, FEMA regulatory context, and the limited legal enforcement options available to Indian lenders for overseas borrowers.

**Time zone compliance**: Kallix's global call scheduling adjusts the 7 AM–7 PM calling window to the borrower's registered country code. A borrower in the UAE (IST – 1.5 hours) receives calls between 8:30 AM and 8:30 PM IST. This is both a compliance requirement (spirit of RBI FPC applies regardless of jurisdiction) and a practical one — calls at 2 AM local time are never answered.

**FEMA implications for NRI defaulters**:
- An NRI in default on a home loan with NRE/NRO accounts at the same lender may have those accounts frozen or offset against the outstanding — lenders have a right of set-off for overdue amounts. The AI communicates this factually.
- Continued NRE account remittances to a lender where a defaulted loan exists can be attached by the lender with RBI permission. NRI borrowers often do not know this.
- OTS settlement: NRIs can pay OTS amounts via inward remittance on the capital account under FEMA. The AI explains the payment mechanism and provides the lender's SWIFT/IBAN details for inward remittance.

**Limited enforcement options**: Indian courts have limited enforcement against NRI borrowers who have no assets in India. The AI does not make threat-based SARFAESI calls to NRI borrowers who have already moved assets abroad — the consequence advisory focuses on what the lender can actually enforce (property in India if secured, NRE/NRO accounts, future India visa restrictions in case of flagging by specific lenders).

**WhatsApp-first for NRIs**: International calls are expensive and time-sensitive. Kallix uses WhatsApp (free for international) as the primary channel for NRI accounts, with voice calls for high-value accounts or OTS discussions.

Microfinance recovery has its own regulatory framework — distinct from standard retail lending — because the borrower profile (low-income, often rural, financially less literate) requires additional protections. RBI Microfinance Directions 2022 (RBI/2022-23/27) are the governing document for NBFC-MFIs.

Key MFI recovery constraints:
- **Group loan context**: Many MFI loans are joint liability group (JLG) loans — the AI must handle individual vs group loan recovery differently. For JLG loans, the AI calls the group leader first if the individual borrower is unreachable.
- **No doorstep visits after 7 PM**: The traditional MFI loan officer visit model is regulated — no collection activity at the borrower's residence after 7 PM. AI calls are also restricted to 7 AM–7 PM for consistency.
- **Maximum one collection contact per week** for the same account in the normal cycle (RBI direction). The AI enforces this weekly cap — maximum 1 AI call per week per borrower at the standard DPD stage.
- **FOIR compliance**: RBI Microfinance Directions 2022 cap total EMI obligations at 50% of household income for MFI borrowers. If a borrower in recovery mentions they have multiple MFI loans, the AI notes this for the lender's credit review team — potential regulatory over-indebtedness flag.
- **Loan card requirement**: MFI borrowers must be provided a loan card showing all charges, interest rates, and the lender's name. If a borrower cannot confirm their loan details, the AI dispatches the loan card details via WhatsApp (in vernacular where configured).
- **Language**: MFI borrowers are typically rural and vernacular-primary — Hindi, Marathi, Tamil, Bengali, Odia are the most common languages. Kallix's regional language support is especially critical for MFI collections.

Quality assurance in debt recovery is a regulatory requirement, not just a commercial one. RBI Recovery Agent Guidelines 2008 require lenders to maintain oversight of all recovery agent conduct — and the same standard applies to AI agents operating on the lender's behalf.

Kallix's QA architecture:

**Automated scoring (100% of calls)**:
- Compliance checklist: lender identity disclosed at start (Y/N), 7 AM–7 PM window confirmed (Y/N), grievance pathway offered (Y/N), third-party disclosure avoided (Y/N), distress signal response correct (Y/N for flagged calls)
- Interaction quality: PTP captured with date and amount (Y/N), hardship profile correctly classified (Y/N), OTS offer presented per eligibility flag (Y/N), consequence advisory accurate (Y/N)
- Call outcome accuracy: LMS write-back matches call transcript (sampled 10%)

**Threshold alerting**: Any call scoring below the lender's configured compliance threshold (default: 90%) generates an automated alert to the lender's compliance manager within 4 hours.

**Weekly human sampling**: 5% of calls (random stratified by DPD stage and script branch) reviewed by a Kallix quality analyst and the lender's compliance team jointly. Monthly report submitted to lender's Collections Head.

**RBI inspection readiness**: All call recordings are stored for 90 days minimum (per RBI DLG 2022). The Kallix compliance dashboard provides RBI inspection-ready exports: call logs, transcript summaries, compliance scores, and exception reports — formatted for presentation to RBI inspectors.

**Script version control**: Every script change is versioned with a timestamp, approver name, and reason for change. RBI inspectors can review the exact script version that was active during any specific call period.

The supervisor dashboard is the human oversight layer that keeps AI recovery within policy and allows rapid exception handling without requiring technical changes to the system.

Key supervisor functions:

**Account-level overrides**:
- Daily call limit override: bump from 2 to 3 calls for a specific account (e.g., PTP broken twice in 3 days)
- Call suppression: pause all calls for an account for 1–30 days (e.g., account under legal dispute, borrower hospitalised, hardship review in progress)
- Script variant assignment: manually assign a specific script variant (e.g., OTS-only, SARFAESI advisory, compassionate mode) to an account
- Manual priority flag: mark account as priority for same-day human callback

**Queue management**:
- AI-escalated queue: all accounts escalated by AI (distress, dispute, sub-threshold payment, hardship) displayed in priority order with escalation reason, time in queue, and transcript summary
- PTP broken queue: accounts where PTP was promised but not honoured, sorted by amount and DPD
- Hot accounts: accounts at 85+ DPD approaching NPA cutoff — displayed as urgent

**Reporting**:
- Daily metrics: calls made, contact rate, PTP captured, PTP kept, payments received, escalations
- Agent vs AI comparison: how AI-led accounts are performing vs human-led accounts in same DPD band
- Script performance: which script branches have highest PTP rate, which trigger most escalations

**Exception log**: Every supervisor override is logged with timestamp, supervisor ID, and reason. This audit trail satisfies RBI's requirement for documented human oversight of recovery agent activity.

The RBI Integrated Ombudsman Scheme 2021 provides borrowers a zero-cost grievance mechanism against banks and NBFCs for service deficiencies, unfair practices, and recovery harassment. Filing an Ombudsman complaint does not automatically stay the debt — the borrower still owes the outstanding — but it places the lender in a formal regulated grievance process where conduct is under scrutiny.

Kallix's Ombudsman complaint handling:
1. **Borrower declaration during call**: If a borrower says 'I've filed an Ombudsman complaint' or 'I've complained to RBI,' the AI immediately stops: 'I've noted your complaint with the Ombudsman. I'm pausing automated collection calls and flagging this for our nodal officer, who will respond through the formal grievance channel. Is there a reference number for your complaint I can note?' Collection suspended pending lender's legal team review.
2. **LMS flag**: 'Ombudsman complaint — collection suspended' flag raised. Human nodal officer assigned within 4 hours.
3. **Collection resumption**: Collection calls may resume only after the lender's compliance team confirms the complaint has been resolved or closed. Kallix does not automatically resume — resumption requires explicit LMS flag update by the nodal officer.
4. **False Ombudsman claims**: Some borrowers mention 'RBI complaint' as a stalling tactic without actually filing. The AI does not verify — it treats all Ombudsman declarations at face value and suspends. The lender's team can verify through the CMS portal and resume if the complaint does not exist. The cost of wrongly suspending is far lower than the cost of a regulatory complaint for continuing calls during a genuine Ombudsman case.
5. **Regulatory risk**: Continuing recovery calls after a documented Ombudsman complaint can result in RBI direction to the lender, financial penalties, and adverse findings in annual inspection reports.

Agricultural loan recovery has a distinct regulatory and humanitarian dimension. The RBI regularly issues guidelines following natural calamities — floods, droughts, cyclones — requiring banks and NBFCs to restructure agricultural loans in affected regions without treating the account as downgraded. Continuing standard recovery calls in affected areas is both commercially counterproductive (the borrower genuinely cannot pay) and reputationally damaging.

Kallix's agricultural loan calamity protocol:

**Geo-tagging and pin code suppression**: When a calamity or crop failure is declared (by state government or RBI circular), the lender uploads affected district pin codes to the LMS. Kallix reads this geo-tag at the account level — all agricultural loan accounts with registered addresses in affected pin codes are automatically flagged and removed from the standard recovery queue.

**NABARD/RBI relief advisory**: The AI, when reaching an agricultural borrower in an affected area before the pin code suppression is applied, is configured to ask: 'Has your district been declared a calamity or drought area? If yes, you may be eligible for loan restructuring under [Bank Name]'s relief scheme. I'll connect you with our agricultural credit team.' This proactive surfacing often reduces delinquency before it deepens.

**Kisan Credit Card (KCC) accounts**: KCC revolving credit accounts have a different repayment structure — the annual renewal and sub-limit enhancement depend on good repayment history. The AI explains that restructuring KCC accounts during calamity relief is possible without affecting the account's KCC renewal eligibility, which is a significant motivator for farmers who rely on annual credit access for the next crop cycle.

**Post-calamity recovery**: Once the relief period (typically 12–24 months) expires, accounts that were suppressed during calamity are reactivated in the recovery queue. The AI uses a 'post-relief reactivation' script that acknowledges the calamity period, states the new outstanding (restructured), and helps the borrower understand the revised repayment schedule.

Credit card debt is structurally the most expensive consumer debt in India — revolving interest at 36–42% APR compounds monthly, meaning a Rs 50,000 balance unpaid for 12 months becomes Rs 68,000–Rs 71,000 even with minimum payments. Borrowers often do not understand this dynamic, and the AI's role in explaining it is both an educational service and a highly effective recovery motivator.

Kallix's credit card recovery configuration:

**Outstanding growth explanation**: 'Your current outstanding is Rs 58,200. At your card's monthly interest rate of 3.4%, this increases by approximately Rs 1,979 each month if no payment is made. Resolving this now saves you Rs 23,748 in interest over the next 12 months.'

**Minimum due trap advisory**: 'Paying only the minimum due (Rs 2,500/month) will keep your account from going further overdue, but at 3.4% monthly interest, it will take approximately 34 months to clear the balance, and you will pay Rs 26,000 in additional interest. A one-time settlement today of Rs 48,000 closes the account at a discount.'

**Settlement offer for credit cards**: Credit card OTS is typically structured as a percentage of total outstanding: 70–85% for 31–60 DPD, 60–75% for 61–90 DPD, 50–65% for NPA accounts. The lender configures these ranges; the AI presents the pre-approved offer amount without revealing the negotiation band.

**CIBIL write-off advisory**: Credit card accounts written off after 180+ days delinquency are reported as 'Written Off' — worse than 'Settled' on the bureau report. The AI uses this urgency to drive OTS before write-off: 'Settling now results in 'Settled' status. After 180 days, the account will be written off and reported as 'Written Off,' which is harder to recover from on your credit record.'

**Card block and limit reinstatement**: Some borrowers in collections are motivated by card reinstatement — 'if I pay the outstanding, will my card be reactivated?' The AI confirms that payment of outstanding + penalties results in card reactivation subject to the lender's review, and that reactivation is significantly faster than applying for a new card from scratch.