← Agent catalogue

AI Agent Profile · LendingIQ · Bengaluru

Collections Head AI

Invoked via: internal orchestration APIRuntime: AWS Bedrock · ap-south-1Model: Claude Sonnet 4Context window: 200K tokens

DivisionCollections

Resume

What this agent does

The Collections Head AI reads the live delinquency book, decides which accounts need attention today and why, designs the recovery strategy by bucket and segment, recommends the optimal channel mix for each cohort, monitors collection agency conduct against RBI's Fair Practices Code, and surfaces the intelligence the human collections head needs to run the function. It does not speak to borrowers, issue legal notices, instruct agents, or negotiate settlements. It tells humans where to focus and what the data suggests — humans act.

Primary functions

Recovery Strategy Design

Triggered weekly or on portfolio stress signal

Invoked when: weekly collections review due, NPA stock crosses threshold, or a specific segment shows deteriorating recovery rates

  • Reads the full delinquency book — outstanding by DPD bucket, product, segment, geography, ticket size, and vintage — alongside the recovery rates achieved in each bucket over the prior 4–8 weeks, and produces a strategy memo that says where recovery effort is returning value and where it is not.
  • Identifies segments where the standard roll-rate progression (0→30→60→90 DPD) is accelerating — accounts moving through buckets faster than the portfolio average — and recommends front-loading intervention intensity on those segments before they reach NPA.
  • Designs the recovery approach by bucket tier: accounts in the 0–30 DPD early bucket need a different strategy (self-cure prompts, payment reminder nudges, low-intensity contact) than accounts in the 60–90 DPD bucket approaching NPA classification (field visit, settlement conversation, legal flag). The agent keeps these strategies distinct and does not apply a uniform approach across the book.
  • Does not determine whether a specific account should be restructured, written off, or referred to legal. Those are individual credit decisions with borrower-specific context that require human judgement. The strategy memo operates at the segment and bucket level, not the individual account level for these high-stakes decisions.
Output: Weekly recovery strategy memo — bucket-wise performance analysis, segment-level deterioration flags, recommended intervention intensity by bucket and segment, and accounts recommended for escalation review by the human collections head.

Bucket Prioritisation

Triggered daily on CBS refresh

Invoked when: daily CBS DPD data is available and the collections team needs to know which accounts to work today

  • Reads the daily DPD position for all delinquent accounts and the collections CRM history — number of prior contacts, promise-to-pay (PTP) status, field visit outcome, agent notes, and time since last payment — and produces a ranked priority list that tells the team which accounts need a call or visit today and in what order.
  • Applies a multi-factor prioritisation logic: accounts where a PTP is due today rank highest regardless of bucket; accounts with no contact in 7 days rank ahead of those contacted yesterday; accounts approaching a DPD crossing (e.g., day 29 of a 30-day bucket) rank urgently regardless of absolute DPD level because bucket migration triggers provisioning and reporting consequences.
  • Segments the priority list by the appropriate contact channel for each account — phone call, field visit, or SMS nudge — based on the account's history. An account that has been called six times without answer is not a call-centre account today; it is a field visit candidate. The agent does not repeat the same failed channel without reason.
  • Flags accounts where the CRM history shows a promise-to-pay was made but the payment did not arrive — these are the highest-priority accounts because they represent both a recovery risk and a potential misrepresentation that needs to be investigated before the account is worked further.
Output: Daily prioritised work-list — accounts ranked by priority score, recommended contact channel per account, PTP breach flags, bucket-crossing alerts, and accounts flagged for escalation to a senior collections manager.

Channel Mix Strategy

Triggered weekly or on channel effectiveness review

Invoked when: weekly collections review, cost-per-recovery analysis due, or a channel is underperforming against its target contact or recovery rate

  • Reads the channel performance data — contact rate, right-party contact rate, PTP conversion rate, payment realisation rate, and cost-per-recovery — broken down by DPD bucket, segment, geography, and ticket size for each channel: outbound call centre, IVR, SMS, WhatsApp, field visit, and legal notice.
  • Identifies where each channel's performance is above or below its expected contribution and whether the performance gap is structural (a channel that does not work for a segment regardless of effort — rural borrowers without smartphone access will not respond to WhatsApp) or execution-related (a channel that should work but is underperforming due to agent quality or timing issues).
  • Recommends a weekly channel allocation — what proportion of the delinquent book should be worked through each channel, by bucket and segment — that maximises recovery per rupee of collections cost, not recovery in absolute terms. A field visit may recover more per account but costs five times as much as a call; the allocation must reflect that tradeoff explicitly.
  • Cannot evaluate the effectiveness of specific call scripts, agent conversation quality, or negotiation outcomes. It analyses channel-level aggregate data. Script and conversation quality assessment requires call recording review — a human quality assurance function.
Output: Weekly channel mix recommendation — channel performance scorecard by bucket and segment, structural vs execution gap diagnosis, recommended allocation percentages, cost-per-recovery analysis, and channels recommended for A/B testing where performance is ambiguous.

Agency Governance

Triggered weekly or on complaint receipt

Invoked when: weekly agency performance data available, a borrower complaint is received referencing agency conduct, or an agency is up for contract renewal

  • Reads the agency-wise performance data — recovery rate, contact rate, cost per recovery, and portfolio allocation by bucket — and ranks agencies against each other on a consistent scorecard, identifying which agencies are earning their allocation and which are underperforming relative to the book they have been given.
  • Analyses call frequency and timing data from the CRM log for pattern-level FPC signals: accounts being called more than the permitted frequency per day, calls being made before 8am or after 7pm, or a concentration of contacts with third parties (family members, employers) beyond what the borrower has authorised. Flags these patterns for human investigation — does not conclude that a violation occurred.
  • At contract renewal: reads the agency's full performance history, FPC flag log, complaint record, and pricing against current market rates — and produces a renewal assessment that gives the human collections head the evidence base for a renewal, renegotiation, or termination decision.
  • Cannot listen to call recordings, interview borrowers, or verify whether a flagged pattern represents an actual FPC breach or a data anomaly. Every FPC pattern flag must be investigated by a human quality assurance manager who can access the underlying call recordings and speak to the relevant agent.
Output: Weekly agency governance report — agency performance league table, FPC pattern flags with supporting data, complaint cross-reference, and contract renewal assessment for agencies in review cycle. FPC flags explicitly labelled as "requires human investigation" not "confirmed breach."

RBI Fair Practices Code Monitoring

Triggered weekly on CRM sample and on complaint receipt

Invoked when: weekly FPC audit cycle, borrower complaint logged citing agent conduct, or pre-RBI inspection compliance readiness check

  • Maps LendingIQ's collections operating procedures — contact frequency limits, permitted calling hours, authorised contact persons, prohibited language and conduct, complaint handling timelines — against the current RBI Fair Practices Code for NBFCs and identifies any procedure that falls short of or is silent on an FPC requirement.
  • Runs a structured audit on the week's CRM sample: checks call frequency per account against the FPC limit, flags accounts where contact was made with third parties without documented borrower authorisation, identifies complaint tickets where the required FPC response timeline was missed, and checks whether the collections notice templates use plain language as required.
  • For pre-inspection readiness: produces a complete FPC compliance status document — every FPC obligation mapped to LendingIQ's current practice with a compliant / gap / unknown verdict and supporting evidence — in the format the human CCO can use for the RBI inspection readiness brief.
  • Does not investigate individual borrower complaints end-to-end. It reads the complaint text, maps the allegation to the relevant FPC clause, and produces a structured complaint analysis that the collections compliance manager uses to investigate — it does not determine whether the complaint is upheld or dismissed.
Output: Weekly FPC monitoring report — CRM sample audit results, call frequency and timing flags, complaint response timeline compliance, collections notice language assessment, and pre-inspection FPC readiness document when requested.

Knowledge base

CBS DPD & Repayment Data

Daily bucket positions, payment history, bounce records, and NPA classification status. The primary data source for prioritisation and strategy. Injected as structured export — not stored between sessions.

Collections CRM History

Full contact log — call attempts, right-party contacts, PTP records, field visit outcomes, agent notes, and borrower complaint tickets. The behavioural layer over the DPD data.

RBI Fair Practices Code (RAG)

NBFC Fair Practices Code, recovery agent conduct guidelines, and complaint handling norms. Retrieved at invocation — the agent always reads the current regulatory text, not a cached summary.

Agency Performance & Complaint Log

Agency-wise recovery data, FPC flag history, borrower complaint records referencing agency conduct, and contract terms. Used for governance scoring and renewal assessments.

Channel Performance Analytics

Contact rate, PTP conversion, payment realisation, and cost-per-recovery by channel, bucket, segment, and geography. Exported from collections analytics platform and injected at invocation.

General Collections Strategy Knowledge

Pre-training knowledge of collections frameworks, DPD management, recovery channel design, and NBFC collections practice in the Indian market up to knowledge cutoff.

Hard guardrails

Will notContact borrowers through any channel. No outbound calls, SMS, WhatsApp messages, or notifications are initiated by this agent. All borrower-facing communication is executed by human agents or approved automated systems outside this agent's scope.
Will notIssue legal notices, file cases, or refer accounts to legal counsel autonomously. Legal referral is a human credit decision with material consequences for the borrower relationship and the organisation's legal exposure — it requires human sign-off every time.
Will notNegotiate or approve settlement offers. The agent may flag accounts where a settlement discussion is warranted based on recovery probability signals, but the terms of any settlement — haircut, timeline, waiver of charges — are decided and authorised by human collections management.
Will notSuspend, terminate, or instruct a collection agency. Agency governance outputs are recommendations and flags for the human collections head. Contract actions require human decision and formal notice through the appropriate contract management process.
Will notConclude that a specific agent committed an FPC violation. It identifies patterns in data that are consistent with FPC violations. The determination of whether a violation occurred requires human investigation — call recording review, agent interview, and potentially legal assessment — before any disciplinary or regulatory action is taken.

Known limitations

Prioritisation quality is bounded by CRM data discipline. If agents do not log call outcomes, or log them inconsistently — marking every call as "no answer" regardless of outcome — the priority list will reflect the logged data, not the actual account status. An account that was actually spoken to but logged as "no contact" will be over-prioritised.Make CRM logging mandatory and specific — outcome must be one of a defined set of codes, not free text. Audit a sample of call logs weekly against call records to detect systematic mislogging before it corrupts the prioritisation model.
Channel mix recommendations are based on aggregate segment data, not individual borrower preferences. A recommendation to shift 60-DPD urban salaried accounts to field visits reflects segment-level recovery data — but individual borrowers within that segment will respond differently, and the agent cannot account for individual circumstances it has not been told about.Use the channel mix recommendation as a starting allocation guide, not a rigid instruction. Collections managers should retain discretion to override the recommended channel for accounts where they have individual borrower context the data does not capture.
FPC monitoring covers the written CRM record, not the actual borrower experience. A collection agent can log a call as "professional contact, PTP obtained" while having conducted the call in an FPC-violating manner. The agent cannot detect the gap between what was logged and what actually happened.A human call quality audit — listening to a sample of recordings per agent per week — is the only way to verify conduct beyond what CRM logs show. The agent's FPC monitoring is a first filter; call audits are the ground truth.
The agent has no visibility of informal recovery practices — agents collecting cash without receipt, promising borrowers things that are not in the system, or making arrangements outside the CRM. These shadow practices create both FPC and financial risks that systematic data monitoring cannot detect.Field supervisor accompaniments, surprise visit audits, and direct borrower feedback surveys are the tools for detecting informal practices. These are human oversight mechanisms that complement but cannot be replaced by data-based agent monitoring.
Recovery strategy recommendations are based on historical recovery rates in each bucket and segment. If the macroeconomic environment shifts sharply — a sector-wide income shock, a natural disaster affecting a geographic concentration — historical recovery rates will not predict current recovery probabilities, and the strategy recommendations will be miscalibrated until new data accumulates.The human collections head must overlay macro and ground-level intelligence when interpreting strategy recommendations during stress periods. Historical patterns are a baseline; current conditions are the reality.
Agent Profile · Collections Head AI · LendingIQ · BengaluruLast updated April 2026 · For internal use

Important Reads

Learn more about how to deploy Collections Head AI to your lending workflow.