← Agent catalogue

AI Agent Profile · LendingIQ · Bengaluru

Mid Bucket Agent AI

Function: DPD 31–90 Collections ContactOperating hours: 8 AM – 7 PM hardcodedRuntime: AWS Bedrock · ap-south-1Model: Claude Sonnet 4Context window: 200K tokens

DivisionCollections

Resume

What this agent does

The Mid Bucket Agent AI contacts borrowers in the DPD 31–90 window using a graduated escalating-tone sequence — firm reminders at DPD 31–45, consequence-informing messages at DPD 46–70, and final pre-legal notices at DPD 71–90 — while detecting hardship signals in real time and routing genuine cases to the human restructuring team before they progress to legal action. DPD 31–90 is the last window where automated contact can change the trajectory of an account. The agent executes approved templates with disciplined escalation logic; human officers handle everything that requires judgment, empathy, or terms discussion.

Primary functions

Escalating Tone Logic

DPD 31–90 — three-phase graduated sequence

Invoked when: account appears on the Collections Head AI's daily work-list for mid-bucket contact

  • Phase 1 (DPD 31–45): firm, non-threatening payment reminders citing outstanding amount, days overdue, accruing late charges, and the path to regularisation. Tone is matter-of-fact and consequence-informing — not alarming. Appropriate for borrowers who may have overlooked or deferred a payment and need a clear statement of what is owed and when.
  • Phase 2 (DPD 46–70): consequence-escalating messages referencing credit bureau delinquency reporting already in effect, accumulating recovery charges per the loan agreement, and the serious risk of legal recovery action if the account is not regularised. Tone is firm and urgent — the borrower needs to understand that the situation has become materially worse since DPD 30.
  • Phase 3 (DPD 71–90): final pre-legal notices stating that the account will be referred for legal recovery proceedings within a specified window absent payment or a restructuring arrangement. These messages are the regulatory and legal gateway to the Hard Bucket — their language is reviewed by legal counsel, not just the compliance team.
  • Checks CBS payment status in real time before every contact attempt — no contact is made on an account where payment has been received in the current day's CBS update. Checks the CRM for any open human officer case on the account — automated contact is suspended for any account where a human officer is in active contact.
Output: CRM contact log — phase applied, template, channel, delivery status, borrower response classification, and CBS pre-check result.

Hardship Detection

Applied to all borrower responses in real time

Invoked when: borrower responds to any contact attempt

  • Classifies every response against a hardship taxonomy: direct declarations (job loss, medical emergency, business failure), distress signals (expressions of inability to pay or requests for help), dispute signals (contesting the debt or charges), and payment intent signals (PTP commitments). Each classification has a distinct routing: payment intent to PTP capture; hardship and distress — automated sequence stops immediately, human officer contacted same working day.
  • When hardship is detected, packages the account brief for the human officer: DPD, outstanding amount, payment history since origination, all prior contact attempts, borrower's stated hardship reason, and the available restructuring options under the credit policy for this product type. The human officer does not receive a cold handover — they receive a prepared brief so their first contact with the borrower is substantive.
  • Does not assess whether a stated hardship is genuine — that requires a human conversation. The agent classifies what the borrower said and routes it. The human officer's assessment determines whether restructuring is appropriate or the account progresses to the hard bucket.
Output: Hardship flag with verbatim response, classification type, account brief for human officer, available restructuring policy options, and same-working-day escalation to human collections team.

Restructure Referral

On confirmed hardship or DPD threshold trigger

Invoked when: hardship confirmed by human officer, or account DPD and exposure meets the restructuring referral policy threshold

  • Packages the restructuring referral file: full payment and contact history, hardship reason documented by the human officer, and the policy options available — EMI reduction via tenure extension, moratorium, partial payment arrangement — formatted for the credit officer who will assess the application. The agent does not propose specific terms; it presents the policy menu.
  • Creates the CRM restructuring case with all required data fields populated, reducing data entry burden and ensuring the file is complete before reaching the credit team. The credit officer validates, proposes terms, and obtains borrower agreement — all human-led steps. The agent's role ends when the referral file is created.
  • Monitors the case outcome: if restructuring is approved, the account exits the mid-bucket contact queue. If declined, the account re-enters the mid-bucket sequence at the appropriate DPD step or transfers to the hard-bucket agent depending on the DPD at that point.
Output: CRM restructuring case — complete account history, hardship documentation, policy options note, routing to credit officer's queue, and case status monitoring for re-entry or hard-bucket transfer on decline.

Hard guardrails

Will notContact borrowers outside 8:00 AM – 7:00 PM. Higher DPD urgency creates no exemption from RBI Fair Practices Code contact hour requirements.
Will notNegotiate, approve, or imply any modification to loan terms. Restructuring, moratorium, settlement — all require human credit officer authority. The agent identifies that these conversations need to happen; it does not have them.
Will notContinue automated contact while a human officer has an open active case on the same account. Automated and human contact on the same account simultaneously creates borrower confusion and FPC risk.
Will notUse threatening, coercive, or shaming language. Consequence-informing is permitted; harassment is not. Every template has been reviewed for FPC compliance.

Known limitations

Hardship detection covers stated hardship. Borrowers who are genuinely struggling but respond with deflection ("will pay next week") rather than disclosure will continue to receive automated escalation rather than the human empathy intervention they need. Repeated short-interval PTP breaches are the primary proxy signal for undisclosed hardship.Build a repeated-PTP flag into the work-list: any account with three or more PTP breaches in the mid-bucket window should trigger a human-initiated hardship inquiry call before the next automated escalation.
The tone sequence is DPD-driven but DPD is a blunt instrument. A first-time delinquent account at DPD 55 deserves different treatment from a serial delinquent at DPD 55 — but the agent applies the Phase 2 template to both unless the Collections Head AI's brief specifically flags the account history for a different approach.Ensure the Collections Head AI includes a delinquency history tag (first-time / repeat / chronic) in every mid-bucket account brief, so human review can intercede when the DPD-band template is a poor fit.
Agent Profile · Mid Bucket Agent AI · LendingIQ · BengaluruLast updated April 2026 · For internal use

Important Reads

Learn more about how to deploy Mid Bucket Agent AI to your lending workflow.