← Agent catalogue

AI Agent Profile · LendingIQ · Bengaluru

Early Bucket Caller AI

Function: DPD 0–30 Collections ContactRuntime: AWS Bedrock · ap-south-1Model: Claude Sonnet 4Context window: 200K tokens

DivisionCollections

Resume

What this agent does

The Early Bucket Caller AI contacts borrowers in the 0–30 DPD bucket with soft payment reminders — through WhatsApp, IVR, or SMS — captures promise-to-pay commitments when borrowers indicate they will pay, and logs every contact attempt and outcome to the CRM in real time. It works through approved message templates only. It operates within the RBI Fair Practices Code contact hours without exception. It escalates any borrower who expresses hardship, requests restructuring, or shows signs of genuine financial distress to a human collections officer immediately. It does not negotiate, does not make exceptions, and does not continue contacting a borrower who asks to stop.

Primary functions

Soft Reminders

DPD 0–30 — first contact through graduated sequence

Invoked when: account appears on the Collections Head AI's daily work-list with a contact instruction for the current day

  • Reads the account brief from the Collections Head AI — DPD days, last payment date, outstanding amount, prior contact history from the CRM, and the recommended channel for today's contact — and executes the contact in the specified channel using the pre-approved template for the account's DPD band and contact attempt number. A first WhatsApp reminder for DPD day 3 uses a different tone than a third-attempt IVR call on DPD day 22 — each template is calibrated for the appropriate moment in the collection cycle.
  • Follows a graduated contact sequence: WhatsApp message first (lowest friction, highest open rate, leaves a non-intrusive record), IVR outbound if no response or no payment within 24 hours, SMS as a fallback for borrowers not reachable on WhatsApp or IVR, and human agent escalation when the automated sequence has exhausted its attempts or the account reaches DPD 25 without resolution. The sequence is configured by the Collections Head team and applied consistently — the agent does not skip steps or apply ad-hoc sequences.
  • Checks the CBS payment status in real time before each contact attempt — if the borrower has made a payment since the last CBS update, the contact is cancelled automatically and a payment acknowledgment message is sent instead of a reminder. Contacting a borrower who has already paid is the most avoidable FPC violation in automated collections; the real-time CBS check prevents it.
  • Operates only between 8:00 AM and 7:00 PM as an architectural constraint — not a soft configuration. Any contact job scheduled outside this window is held until 8:00 AM the following day. The agent does not accept overrides to this constraint from any source, including the Collections Head AI's daily work-list.
Output: CRM contact log entry — channel used, template name and version, dispatch timestamp, delivery status (delivered/failed), and payment status check result. If CBS shows payment received before dispatch: contact cancelled, acknowledgment sent, reason logged.

Promise-to-Pay Capture

When borrower responds indicating payment intent

Invoked when: borrower responds to a WhatsApp message or IVR prompt indicating they intend to pay, with or without a specific date

  • When a borrower responds to a WhatsApp message with a payment intent signal — "will pay tomorrow," "paying on the 15th," "I'll transfer this evening" — the agent classifies the response as a Promise-to-Pay, extracts or confirms the committed payment date, and creates a structured PTP record in the CRM: borrower-stated date, amount confirmed (if stated), channel of promise, and timestamp.
  • For IVR interactions: when the borrower selects the "I will pay" DTMF option and provides a payment date (via DTMF date entry or voice capture where the IVR supports it), the agent creates the PTP record from the captured response. Where the IVR does not capture a specific date, the PTP is recorded as "committed, date not specified" and the account is flagged for a follow-up WhatsApp message asking the borrower to confirm their payment date.
  • Sends a PTP confirmation message to the borrower on the channel where the promise was made — confirming the committed date, the amount due, and the payment link or UPI ID. On the committed payment date, sends a same-day reminder at 10:00 AM if payment has not yet been received in the CBS. If payment is still not received by 5:00 PM, flags the PTP as breached and routes the account back to the Collections Head AI's work-list with a PTP breach flag.
  • Does not negotiate the payment date on the borrower's behalf or grant any extension of terms. If a borrower responds requesting more time than a standard PTP date, the agent acknowledges the request and escalates to a human collections officer — it does not approve payment deferrals, partial payments, or restructuring terms of any kind.
Output: PTP record in CRM — committed date, amount (if stated), channel of promise, confirmation message sent timestamp. On PTP day: same-day reminder logged. On PTP breach: breach flag raised to Collections Head AI work-list with full PTP history.

WhatsApp / IVR Channel Management

Continuous — channel selection and response handling

Invoked at each contact attempt — channel selection is dynamic based on prior attempt history and borrower response patterns

  • Manages the WhatsApp contact workflow end to end: dispatch of RBI-compliant, WABA-approved templates, handling of borrower replies within the 24-hour WhatsApp messaging window, classification of replies (payment confirmation, PTP, hardship, dispute, unsubscribe/stop, no response), and appropriate follow-up action for each reply type. Replies that cannot be classified into the configured response categories are escalated to a human collections officer with the full reply text.
  • Manages the IVR outbound call workflow: call initiation, delivery of the pre-recorded script for the applicable DPD band and attempt number, DTMF response capture (1 = I have paid / 2 = I will pay / 3 = I need help / 4 = Dispute / 0 = Speak to agent), and routing based on the borrower's selection. Option 3 (need help) and Option 4 (dispute) route immediately to a human collections officer — the IVR does not attempt to handle hardship or dispute conversations through automated menus.
  • Respects WhatsApp opt-out signals: if a borrower sends "STOP," replies with "Do not contact me," or blocks the messaging number, the agent immediately adds the account to the Do Not Contact list, logs the opt-out in the CRM, and routes the account to a human collections officer for an alternative contact approach review. The opt-out is processed within the same session — there is no lag period during which the agent might send another message after receiving a stop request.
  • Complies with TRAI's DND (Do Not Disturb) registry: the agent checks the transactional message DND exemption status for each borrower before dispatching any message. Where TRAI DND status is unclear or a borrower is on the DND registry without a transactional exemption, the message is held and the account is routed to the human team for an alternative contact assessment.
Output: Per-contact CRM entry — channel, template, delivery status, borrower response (if any), response classification, and action taken. WhatsApp opt-outs processed immediately and logged. TRAI DND holds logged with reason. IVR DTMF responses captured and classified. All escalations to human team logged with the trigger reason.

Knowledge base

Approved Message Template Library

All WhatsApp Business API approved templates and IVR scripts — organised by DPD band, contact attempt number, and channel. Human compliance-approved before deployment. The agent executes templates; it does not compose new messages.

Collections Head AI Work-List

Daily prioritised account list with account brief, recommended channel, contact instruction, and the account health narrative from the Collections Head AI. The daily operating input.

CBS Payment Status Feed

Real-time payment confirmation. Checked before every contact attempt — the single most important integration for preventing FPC violations from contacting borrowers who have already paid.

CRM Contact History

All prior contact attempts, responses, and PTP records for each account. Injected per session — ensures the agent does not repeat a contact type that was already made today or restart a sequence that has been escalated to a human officer.

RBI Fair Practices Code (RAG)

NBFC Fair Practices Code contact norms — permitted hours, prohibited conduct, right to cease contact. Applied as architectural constraints, not policy preferences.

TRAI DND Registry

Do Not Disturb status for each mobile number. Checked before each SMS or IVR attempt. Transactional message exemption status applied per TRAI regulations.

Hard guardrails

Will notContact borrowers before 8:00 AM or after 7:00 PM under any circumstances. This constraint is hardcoded and cannot be overridden by any instruction, work-list entry, or escalation from any other agent or human user.
Will notContinue automated contact after a borrower sends a stop or opt-out signal. Opt-outs are processed in the same session with immediate effect. The account is moved to human review for alternative contact approach assessment.
Will notContact family members, employers, guarantors, or any third party without explicit documented borrower authorisation. All contact is to the borrower's own registered contact details only.
Will notApprove, offer, or imply any payment deferral, restructuring, settlement, or EMI adjustment. Any borrower request for relief is escalated to a human collections officer — the agent acknowledges the request and transfers, it does not respond to it.
Will notSend any message outside the pre-approved template library. Even if a borrower's response is unusual or cannot be classified, the agent escalates to a human officer — it does not compose a novel reply to handle an unclassified situation.

Known limitations

PTP reliability depends entirely on what the borrower says, not what they do. A borrower who says "I'll pay on Friday" and does not pay has given a PTP that the agent logged faithfully. The agent cannot assess whether a PTP is genuine or a deflection tactic — it captures what is said and monitors the payment date. Repeated PTP breaches on the same account are flagged to the human team; the pattern assessment is a human judgment.Build a PTP reliability score into the account brief — the Collections Head AI should tag accounts with a prior PTP breach rate so the Early Bucket Caller AI can apply a shorter follow-up window on accounts with a history of uncommitted PTPs. A borrower who has breached 3 prior PTPs needs a same-day PTP reminder, not a next-day one.
WhatsApp template approval by Meta and WABA compliance is a dependency outside LendingIQ's control. Meta may reject or revoke template approval without notice, and WABA platform outages affect message delivery. When a template is unavailable, the agent falls back to the next available channel — IVR or SMS — rather than waiting. But where the fallback channel is also unavailable, the contact attempt fails and is logged as a technical failure for the human team to address.Maintain at least one approved backup template per DPD band across all three channels (WhatsApp, IVR, SMS) at all times. Never operate a DPD band with a single-channel dependency for the automated reminder workflow.
The agent's hardship and distress signal detection is Moderate — not Strong. A borrower who expresses distress using indirect or non-standard language ("things are very difficult right now," "it's been a rough month") may not trigger the distress classification, and the automated sequence will continue. The agent's language understanding covers common distress phrases in English, Hindi, and major regional languages, but nuanced or indirect distress signals remain a false negative risk.Train the human collections team to review a sample of "no response classified" escalations weekly specifically to catch nuanced distress signals the agent may have missed. Build a feedback loop where human agents tag missed distress signals — these tags improve the distress classification corpus over time.
TRAI DND compliance for WhatsApp is an evolving regulatory area. WhatsApp Business messaging falls under the Telecom Commercial Communications Customer Preference Regulations (TCCCPR), but the interaction between TCCCPR's transactional message category and collections reminders is not uniformly clear across regulators. The agent applies the current interpretation of transactional message exemption — but this is an area where regulatory guidance could change and LendingIQ's compliance team must monitor actively.Include TCCCPR compliance for WhatsApp collections messaging as a standing item in the CCO AI's quarterly regulatory monitoring review. Any regulatory change to the transactional exemption that affects collections messaging requires immediate template and channel configuration review before the next daily run.
The agent has no awareness of the borrower's emotional state, life circumstances, or the context behind a payment failure beyond what is in the account brief. A borrower who missed a payment due to a family bereavement and a borrower who missed a payment because they chose not to pay receive the same template — because the agent cannot distinguish the two without information it does not have. This is the fundamental limitation of automated early-bucket collections: efficiency at scale comes at the cost of individual sensitivity.Build an opt-in "I have a special circumstance" response option into the IVR and WhatsApp workflows that routes immediately to a human officer with empathy-first training — before the automated sequence makes additional contact. This gives borrowers with legitimate extenuating circumstances a path to human support without requiring them to use distress language that the agent may or may not classify correctly.
Agent Profile · Early Bucket Caller AI · LendingIQ · BengaluruLast updated April 2026 · For internal use

Important Reads

Learn more about how to deploy Early Bucket Caller AI to your lending workflow.