Traditional EMI collection relies on human agents making high-volume outbound calls — expensive, inconsistent, and constrained by calling hours and headcount. Kallix AI voice agents replace human effort across the entire pre-delinquency and early-delinquency cycle without sacrificing empathy or compliance.

The workflow has three stages: (1) Pre-debit reminders — automated calls 3–5 days before EMI due date confirming account balance, NACH mandate status, and due amount; (2) Bounce follow-ups — calls triggered within 2–4 hours of a NACH/ECS return, advising re-presentation or alternate payment; (3) Soft collection calls — structured calls for 1–30 DPD accounts capturing Promise-to-Pay, explaining CIBIL impact, and offering restructuring pathways within RBI Fair Practices Code limits.

All interactions comply with RBI Recovery Agent Guidelines 2008 (no harassment, no third-party disclosure), TRAI TCCCPR 2018 (transactional classification for DND bypass), and the RBI Fair Practices Code for NBFC/banks. Every call outcome — PTP date, payment amount, restructuring request, or escalation flag — is logged to the LMS/CRM in real time.

Kallix customers in the retail lending segment report 68–76% first-call resolution on pre-delinquency reminders and a 22–35% reduction in 0–30 DPD bucket flow-through to 31+ DPD within 90 days of deployment.

The pre-debit reminder is the highest-ROI touchpoint in collections automation — a call placed before any payment failure that prevents the bounce entirely. Kallix pulls the EMI schedule directly from the LMS (Nucleus FinnOne, Newgen, Salesforce FSC, or custom CBS) and triggers outbound calls 3–5 days prior, with a follow-up call 24 hours before due date for borrowers who did not confirm balance.

Call script structure: (1) Identity confirmation using date of birth + last 4 digits of loan account number — no OTP required for outbound reminder calls; (2) EMI details stated clearly: due date, amount, loan account number suffix; (3) NACH bank account confirmation — 'Your auto-debit is set up from your SBI account ending 4521. Please ensure Rs 18,450 is available on 25 May'; (4) Payment alternatives — UPI link via SMS/WhatsApp for borrowers preferring manual payment; (5) Escalation offer — press 1 or say 'connect me' to speak with a loan servicing officer.

For borrowers who request a date change or flag a cash flow issue during the reminder call, the agent captures the request and logs it as a 'pre-delinquency advisory' in the LMS, flagging for a human follow-up within 24 hours. This early capture reduces the 0 DPD to 1–30 DPD roll rate by an average of 18–24% in production deployments.

NACH/ECS bounces are the single largest driver of early-bucket delinquency in retail lending. The speed of follow-up is critical — borrowers who are contacted within 4 hours of a bounce have a 3x higher same-day payment rate than those contacted the next business day (Kallix production data, anonymised, 80+ lender customers).

Kallix listens to bounce events via LMS webhook or daily NPCI NACH return file (uploaded by 11:30 AM per NPCI settlement cycle). The AI classifies the bounce by reason code and tailors the response:

- **R01 — Insufficient Funds**: 'Your EMI of Rs 12,000 due today could not be deducted as your account did not have sufficient balance. You can pay now via UPI — I'll send a payment link to your registered mobile.'
- **R05 — Mandate Frozen / Hold on Account**: Routes to escalation immediately — mandate freeze often indicates a legal hold, regulatory action, or bank-side issue the borrower may not be aware of.
- **R28 — Account Closed**: Prompts borrower to update their bank account and NACH mandate — the agent initiates a mandate amendment flow and logs it.

For standard R01 bounces, the AI offers three payment paths: UPI link via SMS, IMPS transfer details, or branch/ATM deposit instruction. If borrower commits to a payment date and time, the PTP is logged to the LMS with a follow-up call scheduled 2 hours before the promised payment window.

The 1–30 DPD window is the highest-leverage point in collections — most borrowers in this bucket have cash flow timing issues, not genuine credit distress. An empathetic, informative call at this stage resolves 65–72% of accounts without human intervention (Kallix benchmark, retail lending cohort).

Call structure for soft collections:
1. **Identity verification**: date of birth + loan account last 4 digits + registered mobile OTP for sensitive data access.
2. **Overdue statement**: 'Your EMI of Rs 15,500 due on 10 May is now 8 days overdue. Your total outstanding is Rs 15,500 plus late charges of Rs 310.'
3. **CIBIL consequence**: 'As per RBI guidelines, missed payments beyond 30 days are reported to credit bureaus including CIBIL, which can reduce your credit score by 50–100 points and affect future loan eligibility.'
4. **Resolution options**: pay now via UPI (link via SMS), commit to a PTP date with a follow-up reminder, or request restructuring advisory (transferred to a loan servicing officer).
5. **PTP capture**: if borrower commits, the agent confirms the date, amount, and payment method; logs to LMS; schedules a reminder call 24 hours before and on the PTP date.

For borrowers with 20+ DPD, the AI also mentions the legal recovery process pathway — not as a threat, but as factual information per RBI Fair Practices Code: 'If the account remains overdue beyond 90 days, it is classified as an NPA, which may initiate recovery proceedings under the SARFAESI Act for secured loans.'

The regulatory framework governing collection calls in India has three primary pillars, all of which Kallix addresses by design:

**1. RBI Fair Practices Code (updated 2015, applicable to all banks and NBFCs)**: Calls must not be harassing, threatening, or abusive. No contact with relatives, neighbours, or employers except to obtain contact information. No calls before 7 AM or after 7 PM. Kallix enforces a hard 7 AM–7 PM calling window at the telephony layer, not just the script layer — the call physically cannot be placed outside these hours.

**2. RBI Recovery Agent Guidelines 2008**: Recovery agents (including AI systems acting as agents) must identify themselves and the lender at the start of every call. Kallix scripts begin: 'This is an automated call from [Lender Name] regarding your loan account ending [XXXX].' The agent name is disclosed, the purpose is stated, and no deceptive identity is used.

**3. TRAI TCCCPR 2018 and DND Registry**: Collection calls classified as 'transactional' (relating to an existing customer relationship — a loan account) are exempt from DND restrictions. Kallix registers headers with TRAI under the transactional category. However, calls are still limited to 3 per day per unique mobile number across all commercial entities per TRAI regulations.

Additionally, all call recordings are retained for 90 days minimum (per RBI Digital Lending Guidelines 2022 audit trail requirements) and are available for inspection by the lender's compliance team or RBI inspectors.

Promise-to-Pay (PTP) capture is the primary measurable output of a soft collection call. A PTP converts an overdue account from 'unreachable' or 'unresolved' into an active commitment with a defined follow-up cadence. Kallix tracks PTP kept rates (PKR) and PTP broken rates as LMS performance metrics.

The PTP flow within the Kallix agent:
1. AI presents total overdue: principal EMI + applicable late charges + bounce charges.
2. AI offers three commitment options: full payment by a date, partial payment (if the lender's policy permits part-payment), or restructuring referral.
3. Borrower states a date. AI confirms: 'So you're saying you'll pay Rs 14,200 by 25 May — is that right?' The double-confirmation is deliberately phrased as a yes/no question to make the commitment explicit and auditable.
4. LMS updated: PTP record created with borrower name, loan account, promised amount, promised date, call recording reference, and agent session ID.
5. Reminder campaign: SMS 48 hours before PTP date + automated reminder call 24 hours before + day-of call 2 hours before promised time.

If the PTP is broken (payment not received within 4 hours of promised time), the account auto-escalates to the next call cadence and a flag is raised in the LMS for human review. Broken PTPs are tracked separately — accounts with 2+ broken PTPs are escalated to human collectors regardless of DPD count, as they represent a higher behavioural risk signal.

NACH return codes are classified by NPCI into three categories: account-side issues (R01, R28, R29), mandate-side issues (R02, R05), and technical returns (R30 series). Kallix maps each code to a specific conversational script and LMS action:

**R01 — Insufficient Funds (most common, ~60% of bounces)**: The AI informs the borrower, explains late charges accruing, and immediately offers UPI payment via a link sent to the registered mobile. Late charges are calculated in real time from the LMS and stated explicitly.

**R02 — Mandate Invalid / Account Number Mismatch**: The AI asks the borrower to verify their account number and triggers a mandate amendment request. A new NACH registration link (via digilocker-authenticated eSign flow) is sent via WhatsApp.

**R05 — Mandate Frozen or Blocked**: This often indicates a garnishee order, bank regulatory hold, or court attachment — sensitive territory. The AI does not discuss the block's cause, and the call is immediately routed to a human collections officer.

**R28 — Account Closed**: The AI informs the borrower their registered account no longer exists and initiates a bank account update + new NACH registration flow.

**R29 — Payment Stopped by Account Holder**: Borrower has actively instructed their bank to stop the mandate. The AI captures reason, offers restructuring if financial hardship is stated, and escalates to a collections officer for negotiation.

**R30 — Bank Technical Return**: Treated as a temporary failure; LMS schedules a re-presentation on the next business day, and the AI calls to inform the borrower and confirm they'll keep the account funded.

Credit score impact communication is one of the most effective motivators in soft collections — and also one of the most regulated. Kallix scripts are designed to be accurate, factual, and proportionate — never threatening or misleading.

What the AI communicates accurately:
- RBI Master Direction on Credit Information Companies (CICs) requires banks and NBFCs to report credit data to all four bureaus (CIBIL/TransUnion, Experian, CRIF High Mark, Equifax) on a monthly basis.
- A payment missed beyond 30 DPD is reported as 'SMA-1' (Special Mention Account) by banks, which begins affecting bureau scores.
- NPA classification at 90+ DPD triggers a 'default' flag visible to all lenders when they pull the bureau report.
- Once an NPA is reported, it remains on the credit record for 7 years from the date of settlement, even after repayment.
- Score recovery after an NPA can take 12–24 months of consistent on-time payments.

What the AI does NOT say:
- It does not threaten legal action specifically ('we will take you to court') — this is permissible only if action is genuinely being initiated.
- It does not exaggerate score impact ('your score will drop to zero').
- It does not imply the borrower is a fraud or criminal.

For borrowers who are unaware their score has already been impacted, the AI offers to connect them to a loan servicing officer who can confirm current bureau status and advise on score improvement steps.

Restructuring advisory is one of the highest-value functions an AI collection agent can perform — it turns a potential NPA into a retained, performing asset. Kallix supports this through a configurable restructuring offer matrix that lenders upload to the platform based on their credit risk policy.

The restructuring matrix typically includes:
- **EMI deferral**: 1–3 months deferral with interest capitalisation (available for customers with clean pre-hardship payment history)
- **Tenure extension**: Reduction of EMI amount by extending loan tenure, within maximum permitted tenor for the product
- **Step-down EMI**: Temporarily reduced EMI (30–50% of original) for 3–6 months, stepping back up — used for contractual income disruptions like job transition
- **One-Time Settlement (OTS)**: For 60+ DPD accounts, a pre-approved OTS offer from the credit committee, presented only to qualifying accounts as defined in the lender's policy
- **Overdraft conversion**: For eligible borrowers, conversion of the outstanding to an overdraft facility — more flexible but typically higher rate

The AI does not make credit decisions — it reads from the lender's pre-approved offer matrix. It presents the available options, captures the borrower's expressed preference, and schedules a callback from a loan modification officer within 4 business hours. The conversation transcript and the borrower's stated preference are pre-loaded into the officer's CRM screen before the call, reducing average handle time by 40–50%.

RBI Digital Lending Guidelines 2022 require that restructuring offers be presented with a Key Facts Statement (KFS) — Kallix triggers an automated KFS dispatch via WhatsApp/email at the moment of restructuring expression of interest.

The escalation boundary at 31 DPD is both a business decision and a regulatory one. RBI Recovery Agent Guidelines 2008 require that recovery agents (human or automated) for accounts beyond certain stages are specifically authorised and trained — the regulatory intent is to ensure proportionate and compliant conduct as collection intensity increases.

Kallix's escalation architecture:
- **31 DPD threshold**: account auto-escalates in LMS; AI moves to a supportive role (reminders, UPI links, statement dispatch) while a human collector takes the lead.
- **Immediate escalation triggers** (regardless of DPD): borrower requests to speak with a human; borrower mentions dispute, legal notice, or insolvency; borrower uses threatening or abusive language; call classified as distress (death in family, medical emergency, natural disaster — the AI is trained to recognise these signals and pause collection activity).
- **Handover packet**: the LMS records transferred include call history, all PTPs (kept and broken), payment amounts received, NACH return reasons, and the transcript of the most recent AI interaction. The human collector's CRM screen is pre-populated before they call.
- **Parallel AI support**: even after human escalation, Kallix continues sending automated payment nudges via WhatsApp and SMS (not voice calls) so the human collector's effort is amplified rather than replaced.

For banks operating an internal collections team, Kallix integrates with dialers (Genesys, Avaya, Exotel, Knowlarity) so the escalation creates a predictive dialler task automatically.

LMS integration depth is what separates a real EMI collection AI from a generic call bot. Kallix is designed to both read from and write back to the LMS — not just dial and talk.

**Read operations** (LMS → Kallix):
- Daily EMI schedule pull: due dates, amounts, account IDs, borrower mobile
- NACH/ECS return file: uploaded by lender or fetched via NPCI return portal
- Existing PTP records: avoids duplicate collection calls on accounts with live PTPs
- Lender's restructuring matrix: eligibility criteria and offer parameters
- DPD count and collection stage: determines script variant and escalation threshold

**Write operations** (Kallix → LMS):
- PTP created: amount, date, channel commitment (UPI/cash/NEFT), call recording URL
- PTP broken: flag raised with timestamp and call outcome summary
- Bounce follow-up outcome: mandate amendment initiated Y/N, payment collected Y/N
- Restructuring expression of interest: product type, borrower-stated preference
- Call disposition codes: reached/not-reached/busy/wrong-number/deceased — matching the lender's own code taxonomy

**Supported integration methods**: REST API (preferred, real-time), SFTP file exchange (for legacy CBS), and Salesforce native connector (for Salesforce FSC users). Kallix also supports Oracle FLEXCUBE's collection module API and Nucleus FinnOne's NBFC API suite.

Deployment timeline for LMS integration: 3–5 weeks for REST API integration with a modern LMS; 6–8 weeks for legacy CBS with file-based exchange. The integration layer is built and maintained by Kallix — the lender's IT team only needs to provide API credentials and sign off on test results.

Call frequency governance in Indian collections is shaped by two overlapping regulatory frameworks — and Kallix enforces both at the telephony layer, not just in the script.

**TRAI TCCCPR 2018 (Telecom Commercial Communications Customer Preference Regulations)**: Commercial entities (including lenders) may make a maximum of 3 transactional calls per day per mobile number. All calls beyond 3/day to a DND-registered number are non-compliant regardless of transactional classification.

**RBI Fair Practices Code**: No calls before 7 AM or after 7 PM. Calls must not be harassing or repetitive in a way that amounts to psychological coercion. The spirit of the guideline is that frequency itself can constitute harassment if disproportionate.

**Kallix default configuration**:
- Maximum 2 AI collection calls per day per borrower (1 reserved for human collector)
- Minimum 4-hour interval between calls
- 7 AM–7 PM window enforced at telephony routing layer
- Accounts on DND: calls placed as transactional (borrower opted in to loan communications), but TRAI header must be correctly registered under transactional sender ID
- Supervisor override: up to 3 total calls/day for specific high-risk accounts (31+ DPD, PTP broken 2x)
- Holiday/Sunday calls: permitted for collection (RBI FPC does not exempt weekends for overdue accounts), but Kallix default is to reduce to 1 call on Sundays and public holidays as a best-practice configuration

All call attempts, including no-answer and busy, are logged per TRAI's record-keeping mandate for commercial communications.

Language mismatches are one of the most significant causes of collection call failure in India — a Kannada-speaking borrower receiving a Hindi collection call will typically drop the call within 10 seconds. Kallix addresses this through automatic language selection at the point of call initiation.

Language determination logic:
1. **Registered state mapping**: borrower's address state maps to primary language (e.g., Tamil Nadu → Tamil, Gujarat → Gujarati).
2. **Prior call history**: if a previous call detected a language preference via keyword signals, that preference is stored in the borrower profile.
3. **Spoken word detection**: if the borrower responds in a different language than the opening, the AI detects and switches within one exchange.
4. **Manual override**: the borrower can say 'Hindi mein baat karo' or 'speak in English' and the AI switches immediately.

For collections specifically, tone calibration differs by language. The Kallix Hindi/Hinglish collection scripts are calibrated for directness without aggression ('Aapka EMI 8 din se due hai — kya aaj payment kar sakte hain?'). Tamil and Telugu scripts are calibrated for formal register, as informal tone in these languages can be perceived as disrespectful in a financial context.

Regional language support also matters for EMI documentation — the AI can send WhatsApp payment links and reminder messages in the borrower's preferred language, including Devanagari script for Hindi, Tamil script, and Telugu script, which significantly improves response rates in non-English markets.

The economics of AI voice for EMI collections work at virtually any scale, but the break-even analysis changes significantly with portfolio size:

**Per-call economics**:
- AI (Kallix): Rs 15–25/call (includes telephony, AI processing, LMS write-back, recording storage)
- Human collections agent: Rs 80–120/call (salary + overhead + dialler + supervisor cost amortised per contact)
- Net saving: Rs 55–95 per call

**Portfolio-level impact (50,000 monthly active accounts)**:
- Total calls/month (avg 3 per account): 150,000 calls
- Human cost: Rs 1.2–1.8 crore/month
- AI cost: Rs 22.5–37.5 lakh/month
- Monthly saving: Rs 82.5 lakh–Rs 1.43 crore

**Beyond cost — recovery rate impact**:
- Pre-delinquency containment: 18–24% fewer accounts roll from 0 to 1 DPD (Kallix production benchmark)
- Early bucket resolution: 65–72% of 1–30 DPD accounts resolve without human escalation
- PTP kept rate: 54–63% of AI-captured PTPs are honoured within 48 hours
- DPD roll reduction: 22–35% fewer accounts cross 30 DPD into the hard collections bucket

**Deployment cost and payback**:
- Setup: Rs 8–18 lakh one-time (LMS integration + script development + UAT)
- Monthly platform: Rs 2–6 lakh depending on call volume
- Payback period: typically 30–45 days at 50,000+ active accounts

For lenders running dedicated collections teams, Kallix does not replace human collectors — it filters the portfolio so human effort concentrates on 31+ DPD accounts, fraud suspects, and restructuring negotiations, where human judgment adds the most value.

Unreachable borrowers are a persistent challenge in collections — and the handling of no-contact accounts significantly affects NPA outcomes. Kallix distinguishes between three types of unreachability:

**Type 1 — Temporary Unreachability (busy/no-answer/call-dropped)**: The system retries within the same day up to the configured daily limit, spacing calls 4+ hours apart. After 3 no-answers in a day, the account enters a 24-hour cooling period before the next day's attempt cycle.

**Type 2 — Network/Device Unreachability (switched-off/out-of-network)**: Kallix triggers a parallel WhatsApp message and SMS to the registered number so that when the borrower's device reconnects, they have a written message awaiting. Research shows that 61% of borrowers with switched-off phones at time of call pick up a WhatsApp message within 6 hours of reconnecting (Kallix platform data, anonymised).

**Type 3 — Wrong Number / Changed Number**: If a call connects and the person identifies as not the borrower, the agent immediately confirms ('I'm sorry, I may have the wrong number — can I confirm, is this [borrower first name]?'), notes the response, and flags the account for skip tracing in the LMS. The skip tracing workflow includes: alternate number on file, WhatsApp message to the primary number (different delivery mechanism than voice), contact at registered address by field agent, and LinkedIn/email outreach if the lender's policy permits.

All unreachable dispositions are logged with timestamps. TRAI requires commercial communicators to maintain records of all contact attempts — Kallix's logging satisfies this requirement automatically.

Segmentation is the difference between a collections AI that builds customer trust and one that damages it. Kallix's collection calling behaviour is driven by the borrower's delinquency history profile, which is read from the LMS at the start of each call.

**First-time missed payment profile (no prior bounces in 12 months)**:
- Script opens with: 'Hi, this is an automated reminder from [Lender] regarding your loan account ending [XXXX]. Your EMI of Rs 18,500 due on 10 May has not yet been received. This could be an oversight — you can pay now via UPI, or if there's an issue we can help with, press 1 to connect with our loan servicing team.'
- Tone: low-pressure, solution-first. No consequence discussion unless the borrower brings it up or the call reaches 15+ DPD.
- CIBIL advisory introduced only after 15 DPD for first-time defaulters.

**Moderate repeat profile (1–2 prior bounces, 12-month window)**:
- Script acknowledges the prior history briefly: 'You had a similar delay in November — I understand cash flows can be unpredictable. Can we set up a payment plan to avoid the same outcome this time?'
- Restructuring offer surfaced earlier (7 DPD vs 15 DPD for first-timers).

**High-risk repeat profile (3+ bounces or any prior NPA)**:
- PTP script immediately: 'Your account has a history of payment delays. Today I need a specific commitment — which date and how much can you pay?'
- Consequence summary (CIBIL, NPA) stated at call open.
- Escalation to human at 7 DPD instead of 31 DPD.

The segmentation logic is configurable — lenders can adjust DPD thresholds and profile criteria based on their credit risk policy and portfolio mix.

The deployment is structured to minimise risk to the lender's portfolio — Kallix pilots on a controlled cohort before full rollout, so any script calibration issues are caught before they affect large account volumes.

**Phase 1 — Integration & Data (weeks 1–2)**:
- LMS API or SFTP credentials established
- EMI schedule pull and NACH return file mapping tested
- Borrower data schema mapped: loan account number, mobile, DPD, product type, EMI amount, due date, NACH bank account
- Write-back schema for PTP records, call dispositions, and bounce outcomes verified

**Phase 2 — Script & Compliance Build (weeks 2–4)**:
- Kallix content team builds language-specific scripts for: pre-debit reminder, bounce follow-up (all NACH return codes), 1–7 DPD, 8–15 DPD, 16–30 DPD, PTP capture, restructuring offer, escalation
- Scripts reviewed by lender compliance team against RBI Fair Practices Code, TRAI TCCCPR headers registered
- Language calibration for all required vernacular variants

**Phase 3 — UAT (week 5)**:
- 150+ test scenarios: NACH return code variants, PTP flow variants, identity verification edge cases, language switch, distress signal detection, escalation handover
- Lender QA team validates call recording quality, LMS write-back accuracy, and escalation timing

**Phase 4 — Pilot (week 6–7)**:
- Controlled cohort: typically 2,000–5,000 accounts in the pre-debit and 1–15 DPD bucket
- Metrics tracked: contact rate, PTP rate, payment conversion rate, escalation rate
- Script refinements based on pilot data

**Phase 5 — Full Deployment (week 8)**:
- Full portfolio segment activated
- Lender receives dashboard: daily call volumes, PTP rates, payment conversions, DPD distribution

RBI Digital Lending Guidelines 2022 (RBI/2022-23/111) represent the most significant update to digital lending regulation in India. For AI collection agents, the relevant provisions are:

**LSP Disclosure**: Any entity acting as a Lending Service Provider must identify itself as an LSP at the start of every interaction. Kallix scripts begin: 'This is [Agent Name], a digital service operated by Kallix on behalf of [Lender Name].' This satisfies the disclosure requirement.

**Key Facts Statement (KFS)**: For any interaction involving loan terms — including restructuring offers, rate changes, or new product offers — a KFS must be delivered within 60 seconds to the borrower. Kallix triggers an automated WhatsApp/email KFS at the moment of restructuring expression of interest during a collection call.

**No Automatic Credit Enhancement**: The guidelines prohibit automatic enhancement of credit limits without explicit borrower consent. The Kallix collection agent never offers increased loan amounts or automatic top-ups as a collection incentive.

**Cooling-Off Period**: Borrowers have a 3-day window to exit a new digital loan without penalty. For borrowers in collections who take a new restructured loan offer, the 3-day cooling-off is preserved and communicated.

**FLDG Cap**: First Loss Default Guarantee arrangements between lenders and LSPs are capped at 5% of the loan portfolio. This is a lender-side compliance requirement; Kallix informs lenders of this limit in the commercial agreement.

**Grievance Mechanism**: RBI DLG 2022 requires a visible, accessible grievance officer mechanism. Kallix provides a dedicated escalation path to the lender's nodal officer during every collection call — borrowers can request the grievance officer contact details at any point.

Third-party disclosure of loan information is one of the most serious compliance risks in collections. A collection call that reveals overdue details to a family member, employer, or neighbour — even unintentionally — is a violation of both the RBI Fair Practices Code and the IT Act 2000 (Section 72A, privacy of financial information).

Kallix's identity verification for outbound collection calls is designed to be fast (completed in under 20 seconds) while preventing accidental disclosure:

1. **Opening statement** (before any account information): 'This is an automated message for [Borrower First Name]. If you are not [First Name], please disconnect this call. I will not share any account information until identity is confirmed.'

2. **First factor — Loan account last 4 digits**: 'Please say or press the last 4 digits of your loan account number.' Voice input or DTMF accepted.

3. **Second factor — Date of birth**: 'Please say your date of birth in DD MM YYYY format.' Matched against LMS record.

4. **Single-attempt sensitivity**: If the borrower provides the correct loan account digits but incorrect DOB, the agent does not confirm the loan account digits are correct — it only says 'verification unsuccessful' to prevent progressive guessing.

5. **Third-party identified**: If someone says 'this is his wife, he's not available,' the agent provides only the lender's contact number and asks the borrower to call back — no account details, no DPD count, no amounts.

For high-DPD accounts (30+ DPD), an optional OTP verification layer can be added — the LMS triggers an OTP to the registered mobile number, which the borrower reads back to the agent. This is recommended for restructuring conversations where specific financial commitments are being made.

Cross-sell during a collections call sounds counterintuitive, but the data supports it for current borrowers (0 DPD) with strong repayment history. The Kallix agent follows a conditional script: the top-up offer is only surfaced if (a) the borrower confirms payment intent or states the EMI has already been paid, and (b) the LMS eligibility flag is positive (pre-computed by the lender's underwriting engine).

The agent says something like: 'Great, your EMI is already on its way. By the way, since your repayment record is excellent, you have a pre-approved top-up of Rs 2 lakhs — no paperwork needed. Want me to connect you to our loan team?' The warm transfer or callback scheduling happens in the same call.

This approach works because the borrower is already engaged, the trust signal (existing relationship) is high, and the offer is genuinely pre-approved rather than a prospecting pitch. Lenders using Kallix for EMI reminders have added top-up disbursal pipelines worth Rs 8–20 crore per month from borrower bases of 10,000–50,000 accounts.

Compliance note: the top-up offer must be disclosed as a loan product, and the borrower must be given the option to receive the Key Fact Statement (KFS) before accepting — consistent with RBI Digital Lending Guidelines 2022.

Moratorium and EMI holiday requests spike during financial stress events (job loss, medical emergency, natural disaster) and were broadly permitted during COVID-19 (RBI Circular March 2020). Outside blanket relief periods, moratoriums are product-specific: some home loans allow one or two EMI skips; most personal and vehicle loans do not have a built-in holiday provision.

The Kallix agent is scripted to: (1) acknowledge the borrower's request without raising false hope; (2) check LMS product eligibility — if the loan product supports a holiday, confirm the number of available skips and the interest accrual method; (3) explain clearly that 'deferring your EMI doesn't waive interest — the deferred amount is added to your outstanding principal'; and (4) warm-transfer to a loan officer who can action the request.

For products without moratorium eligibility, the agent pivots to EMI restructuring options: extended tenure, partial prepayment, or hardship waiver requests — all of which require human officer approval.

Lenders deploying Kallix for moratorium advisory report a 40% reduction in borrower escalation complaints related to 'I didn't understand what I agreed to' — the leading cause of post-moratorium disputes.

Credit card EMI conversion is a high-margin product for card issuers — it converts revolving credit (which may be paid in full at month-end) into a fixed-tenure interest-bearing product. The challenge is that most cardholders are unaware of the conversion option or find the in-app flow confusing.

Kallix integrates with the card management system (CMS) via API to receive transaction webhooks. When a transaction above the issuer-set threshold occurs, the agent calls within 48 hours (before the billing cycle closes, maximising conversion eligibility). The agent explains: 'You spent Rs 45,000 at X. You can convert this to 6 EMIs of Rs 7,900, with a processing fee of Rs 299. Would you like to do that now?'

Borrowers who ask about the interest rate or compare with paying in full are handled by the agent's objection script — if the cardholder is unconvinced, the agent logs the decline and schedules a follow-up nudge at the next statement generation date.

For no-cost EMI transactions (merchant-subsidised), the agent still calls to confirm the conversion and ensure the cardholder knows the correct EMI amount and debit date, reducing disputes at statement generation.

Prepayment enquiries are high-intent signals — borrowers asking about foreclosure are often holding a lump sum (bonus, inheritance, property sale proceeds) and will act quickly if the process is frictionless. The delay is usually caused by borrowers not knowing: (a) the exact outstanding principal today, (b) the applicable prepayment penalty, and (c) the interest saved by foreclosing now vs. in 12 months.

The Kallix agent pulls this from the LMS in real time and presents it as: 'Your outstanding principal today is Rs 18.4 lakhs. Since this is a floating-rate home loan, there is no prepayment charge. If you foreclose today, you save Rs 3.2 lakhs in interest compared to running the loan to term. Should I generate a foreclosure statement and send it to your registered email?'

For fixed-rate loans with prepayment charges, the agent discloses the charge percentage, computes the net benefit, and — if the borrower wants to proceed — warm-transfers to a loan officer to initiate the foreclosure process.

RBI mandates: For floating-rate home loans, no prepayment penalty is allowed (Circular 2012). For floating-rate non-home loans, prepayment penalty is permitted for non-individual borrowers only. Fixed-rate loans may carry a penalty, which must be disclosed upfront.

Gold loan auction is the nuclear option for lenders — it recovers outstanding principal but destroys borrower trust and invites regulatory scrutiny. IRDAI and RBI require lenders to give adequate notice before auctioning pledged assets, but 'adequate notice' in practice often means a registered letter that arrives too late for borrowers in rural or semi-urban areas.

Kallix provides a tiered outreach sequence for gold loan maturity:
- **T-30 days**: First reminder call — maturity date, outstanding principal, renewal option and rate, repayment link
- **T-14 days**: Second call — escalated tone, option to partially repay and renew the remainder
- **T-7 days**: Daily calls + WhatsApp — urgent tone, UPI payment link in every message
- **T-3 days and T-1 day**: Maximum urgency, warm transfer to branch manager or DSA offered

For borrowers who cannot repay in full, the agent is scripted to offer partial repayment + top-up renewal (paying down interest + partial principal to roll the loan). This option is disclosed as available and the borrower is connected to the branch to action it.

Lenders with Kallix on gold loan maturity report 60–75% reduction in auction volumes and 20–30% increase in loan renewal rates (vs. borrowers who simply don't renew due to inaction rather than inability to pay).

Vehicle loan repossession is operationally expensive (towing, storage, auction logistics) and legally complex — the Supreme Court mandated proper notice and procedure in cases like ICICI Bank vs. Shanti Devi Sharma. Avoiding repossession is almost always preferable for both lender and borrower.

The key window is 1–30 DPD (soft collections), where borrower intent is still recoverable. Many 30 DPD accounts are there due to NACH bounce (bank account low that month), salary delay, or simple forgetfulness — not genuine inability to pay. Kallix agents call within 24 hours of a missed EMI, establish the reason, and offer an immediate resolution:

- **Forgot / temporary cash flow**: payment link via SMS and WhatsApp, confirm receipt
- **Salary delay**: reschedule payment for a specific date, create PTP record in LMS
- **NACH bounce**: check if borrower wants to pay now via UPI or reschedule NACH re-presentation
- **Genuine hardship**: flag for restructuring advisory and human officer contact

For 31–60 DPD (hard collections), the agent escalates tone and frequency but continues to offer restructuring before the account hits the repossession threshold.

Lenders who deploy Kallix at 1 DPD (same-day intervention) see 28–38% lower roll-through to 60+ DPD compared to lenders who start collections at 7 DPD.

Home loans are structurally different from personal loans for EMI management: the ticket size is large (Rs 20 lakh–2 crore), the tenure is long (15–20 years), the borrower relationship is high-value, and the consequences of default are severe (property auction under SARFAESI 2002). This means the collections approach must be more advisory and less transactional.

Kallix home loan EMI flows:

**1. Pre-debit reminder (T-3 and T-1)**: Standard reminder with account name, EMI amount, and debit date. For home loans, the agent also confirms whether the borrower is on a floating rate and whether a rate change occurred since the last EMI — avoiding the surprise of a higher debit.

**2. NACH bounce recovery**: Same-day call after bounce, but with explicit reassurance tone: 'Your EMI of Rs 34,800 could not be debited — this sometimes happens due to a bank processing delay. Would you like to pay via UPI link now or should we retry the NACH in 3 days?'

**3. Rate change notification (floating-rate loans)**: When RBI changes the repo rate, Kallix triggers an outbound call to all affected borrowers explaining the new EMI amount, the revised tenure impact, and options (accept new EMI, part-prepay to maintain original EMI, extend tenure).

**4. Prepayment advisory**: As described in the foreclosure advisory question — especially relevant for home loan borrowers who received bonuses or property sale proceeds.

Microfinance Joint Liability Groups operate on a peer-lending model where group members are collectively liable for each other's repayments. Collections traditionally happen at weekly centre meetings facilitated by field credit officers (FCOs). AI voice agents add value in the pre-meeting and post-meeting phases, not the meeting itself.

Pre-meeting phase (T-1 day before centre meeting):
- Reminder call to each group member confirming meeting time, location, and EMI amount
- For borrowers who indicate they may miss the meeting, the agent escalates to the FCO and offers a digital payment option

Post-meeting phase (same day or T+1):
- For borrowers flagged as absent or partially paid, the agent calls to understand the reason and capture a PTP
- For borrowers who paid, the agent confirms receipt and provides a WhatsApp confirmation

MFI-specific compliance constraints:
- NBFC-MFI Fair Practices Code (RBI): EMI calls must not involve coercion, must not involve calling family members without consent, and must respect calling hours (8 AM–7 PM)
- RBI MFIN guidelines cap total indebtedness — the agent does not offer top-up products to JLG borrowers without explicit credit bureau check
- Centre leader (group president) can be notified of group-level attendance risk, but individual member payment status cannot be disclosed to other group members (privacy constraint)

Urban MFI portfolios (city-based borrowers with smartphones) see higher AI resolution rates (35–45%) than rural portfolios (20–28%) due to digital payment adoption.

The mechanics of same-call payment link delivery are straightforward but require tight integration: the agent's telephony layer (Kallix cloud infrastructure) triggers a webhook to the payment gateway (Razorpay, Cashfree, or NPCI-certified partner) which generates a short UPI deep link pre-filled with payee VPA, amount, and reference number. The link is dispatched to the borrower's registered mobile number via SMS and WhatsApp simultaneously.

The agent says: 'I have sent you a payment link on your registered number ending in XXXX — it expires in 30 minutes. Would you like a minute to open it while I stay on the line?' This phrasing keeps the borrower on the call during the payment action, which is the key driver of the 34–42% same-call conversion rate.

For borrowers on feature phones without smartphones or WhatsApp, the agent offers a USSD payment option or routes to a human agent who can initiate a debit authorisation over the phone.

Compliance notes:
- UPI AutoPay mandates require explicit borrower consent — the Kallix agent only triggers an AutoPay registration flow after explicit verbal consent, confirmed by OTP
- Dynamic payment links are single-use and expire in 30–60 minutes — no re-use risk
- Payment confirmation is sent back to LMS via webhook, updating DPD status in real time

NPCI UPI AutoPay (circular 2021) allows lenders to set up recurring debit mandates up to Rs 1 lakh without additional authentication per transaction — Kallix agents use this for re-authorisation after a bounce.

Bullet loans are common in working capital financing, gold loans, and some MSME term loans. The entire principal is due on a single date — making that date a high-stakes event. Unlike EMI defaults (which give a lender 30–90 days to act), a missed bullet repayment immediately triggers NPA classification under RBI norms.

Kallix bullet repayment reminder sequence:
- **T-45 days**: First call — maturity date, outstanding principal, renewal options if available
- **T-30 days**: Second call — offer to initiate refinancing or rollover request if borrower needs extended tenure
- **T-14 days**: Third call — escalated tone, confirm fund arrangement plan
- **T-7 days**: Daily calls + WhatsApp — UPI payment link, warm transfer to relationship manager
- **T-1 day**: Final reminder with immediate warm transfer to RM if no confirmation

For overdraft accounts:
- OD renewal typically requires annual review of financial statements and limit reconfirmation
- Kallix agents call 45 days before expiry to initiate document collection (ITR, GST returns, financials)
- The agent schedules a meeting with the relationship manager for the credit review
- Failure to renew on time results in the OD account converting to a term loan, often at a higher rate — the agent explains this consequence clearly

Borrowers with Kallix bullet reminder flows show 78–85% on-time payment vs 55–65% for lenders relying on email reminders alone.

RBI Recovery Agent Guidelines 2008 define a recovery agent as 'any person engaged by a bank for the purpose of recovery of dues from borrowers.' The guidelines require: (a) the agent must be certified by IBA-approved institutes; (b) the agent's identity and the bank's authorisation must be disclosed to the borrower; (c) agents may not resort to any harassment, intimidation, or illegal methods.

AI voice agents are not classified as recovery agents in the traditional sense — they are automated service systems operating on behalf of the lender, similar to IVR systems that have been in use for collections for decades. The RBI has not separately classified AI voice systems as recovery agents (as of 2026), but the spirit of the guidelines — no harassment, no deceptive practices, no third-party disclosure — must be followed, and Kallix builds these constraints into every script.

The practical implication for lenders:
- **1–30 DPD**: Kallix AI handles the entire soft collection cycle. No registered recovery agents required for this bucket.
- **31–90 DPD (hard collections calls)**: Human agents with IBA certification required. Kallix provides the escalation pipeline and account context.
- **90+ DPD (field recovery)**: Registered field recovery agents required, certified per RBI 2008 guidelines. Kallix does not replace field agents.

By reducing roll-through from 0–30 DPD to 31+ DPD by 22–35%, Kallix directly reduces the volume of accounts requiring certified recovery agents, which is the most expensive and capacity-constrained part of the collections stack.