← Agent catalogue

AI Agent Profile · LendingIQ · Bengaluru

Document Ops Agent AI

Function: Document Collection ExecutiveInvoked via: onboarding pipeline + stalled application triggerRuntime: AWS Bedrock · ap-south-1Model: Claude Sonnet 4Context window: 200K tokens

DivisionLending Operations

Resume

What this agent does

The Document Ops Agent AI manages the document collection phase of the loan application journey — running the completeness check against the applicable document checklist as each upload arrives, extracting key fields from uploaded documents for the origination queue, and running a structured nudge sequence when applications stall due to missing documents. It is the operational layer that turns stalled applications back into active pipelines. It collects and checks; the Document Verification Agent AI then checks quality and forgery; the Loan Origination Agent AI processes the complete file.

Primary functions

OCR Extraction

Every uploaded document — synchronous

Invoked when: a borrower uploads a document to the application journey — any document type

  • Reads the uploaded document and extracts the key fields relevant to the document type: for identity documents, name, date of birth, document number, and expiry date; for income documents, income figure, employer name, period covered, and issuing entity; for bank statements, account number, IFSC, account holder name, and statement period. Each extracted field is accompanied by a confidence score.
  • Passes the extracted data to the Document Verification Agent AI for quality and consistency checking, and to the Loan Origination Agent AI's data queue for LOS pre-population. The extraction is the handoff point — this agent extracts; the verification agent validates; the origination agent uses the data.
  • Where OCR confidence on a critical field is below threshold — the document image is too dark, too blurry, or photographed at an angle that makes text unreadable — triggers a re-upload request to the borrower immediately, explaining specifically which field could not be extracted and what image quality would allow extraction. A re-upload request is faster than letting the document sit in the queue and be rejected downstream by the verification agent.
Output: Extracted data JSON per document — key fields with confidence scores. Re-upload request where confidence is below threshold on a critical field, with specific image quality guidance. Extraction output passed to Document Verification Agent AI and Loan Origination Agent AI data queue.

Completeness Check

After each upload and on daily batch for stalled applications

Invoked when: a document is uploaded (incremental check) or daily batch reviews all applications with outstanding document requirements

  • Reads the application's product type, borrower category, and KYC tier from the application record, retrieves the applicable document checklist from the credit policy corpus, and checks each checklist item against the uploaded document set — producing a completeness status: Complete (all mandatory documents received and OCR-extractable), Incomplete (specific mandatory documents missing, listed by name and policy basis), or Deficient (documents received but quality too low for extraction on one or more required items).
  • On completeness confirmation — all mandatory documents present and extractable — signals the Loan Origination Agent AI to proceed with bureau pull and LOS entry. On incomplete or deficient status, initiates the nudge sequence rather than letting the application sit silently in the stalled queue.
  • Logs every completeness check result with timestamp for the CX Strategy Officer AI — the stall point data (which specific document causes the most stalls, at what step in the journey, and in which segment) is the input to journey redesign decisions. Document collection friction that affects 30% of applicants at a specific step is a product problem that this agent's logs make visible.
Output: Completeness status — Complete (pipeline triggered) / Incomplete (missing items listed with policy basis and alternative options) / Deficient (specific quality issues listed). Completeness event logged for CX Strategy Officer AI with stall point metadata.

Reminder Nudges

Structured sequence for stalled applications

Invoked when: an application is incomplete and the stall duration threshold triggers a nudge

  • Runs a three-stage nudge sequence for incomplete applications: Day 1 — "Your application is nearly complete. You are missing [specific document]. Here is what to upload and why we need it." Day 3 — "We still need [specific document]. If you do not have this exact document, here are three alternatives that satisfy the same requirement under our credit policy." Day 7 — "We can help you complete your application. Here is a step-by-step guide for uploading [specific document], and here is our Account Aggregator pathway which may require fewer documents for your income verification." Each nudge is specific, actionable, and tells the borrower what alternatives exist — not just that something is missing.
  • Sends nudges through the borrower's preferred channel — WhatsApp first (highest open rate), email as fallback, SMS as a last resort — using pre-approved templates for each nudge type and document category. Nudge templates are approved by the compliance team to ensure they are clear, not pressuring, and do not make any credit commitment or promise. They are factual: "you are missing X, here is how to provide it."
  • Classifies borrower responses to nudges: document uploaded (check completeness again, remove from stall queue), alternative document requested (route to credit team exception review), help requested (route to Account Executive AI for assisted application), stop/opt-out (remove from nudge sequence, mark application as abandoned by borrower), and no response (proceed to next nudge in sequence until Day 10 flag). No response after Day 10 with no uploads is escalated to the human team for a decision on whether to close the application or attempt a human outreach call.
Output: Nudge dispatch log — template sent, channel, timestamp, delivery status, borrower response classification. Stall resolution status after each nudge (document received / still outstanding / alternative requested / abandoned). Day 10 flag to human team if application remains stalled after full nudge sequence.

Knowledge base

Credit Policy Document Checklist (RAG)

Applicable document requirements by product, borrower category, and KYC tier. Retrieved live — a checklist change takes effect immediately on the next completeness check. The single most important configuration for this agent's accuracy.

Nudge Template Library

Approved nudge templates by document type, nudge stage, and channel. Includes alternative document options per mandatory checklist item — the "if you do not have X, you can submit Y or Z" guidance that makes nudges actionable rather than just reminders.

Alternative Document Map

The policy-approved alternatives for each mandatory document — which substitute documents satisfy the same requirement and under what conditions. Built from the credit policy corpus and updated when the policy changes.

Approved Nudge Template Library

WhatsApp, SMS, and email nudge templates per document type and nudge stage. Compliance-approved before deployment — the agent executes approved templates only.

AA Pathway Documentation

Account Aggregator consent and document pathway — the alternative route for borrowers who find traditional document upload too complex, reducing the document burden for income verification to a single AA consent step.

Document Collection Knowledge

Pre-training knowledge of Indian lending document requirements, KYC document categories, OCR extraction best practices, and borrower-friendly document collection design up to knowledge cutoff.

Hard guardrails

Will notAccept a document as satisfying a checklist requirement where the document type does not match the requirement — even where the OCR extraction succeeded. A utility bill does not satisfy an income proof requirement regardless of how clearly it extracts. Checklist compliance is categorical, not substitutable without credit team approval.
Will notContact borrowers before 8:00 AM or after 7:00 PM for nudge sequences. The FPC contact hours constraint applies to document collection nudges exactly as it applies to collections contact.
Will notAccept an exception document without human credit team approval. Where a borrower indicates they cannot provide a mandatory document and requests an alternative not in the standard alternative map, the agent flags the request to the credit team — it does not approve the exception autonomously.
Will notAssess document authenticity or forgery indicators. Document quality extraction is this agent's function. Whether an extracted document is genuine or forged is the Document Verification Agent AI's function, running in parallel.

Known limitations

OCR accuracy declines sharply with poor-quality uploads from low-end devices. A borrower photographing a bank statement with an entry-level smartphone in poor lighting will produce an image that extracts poorly regardless of extraction model quality. The re-upload request with specific image quality guidance helps — but some borrowers cannot produce a better-quality image with their available device.Build an alternative pathway for applicants whose device quality consistently produces poor OCR results — defaulting to the Account Aggregator consent pathway for bank statement and income verification, which bypasses device-quality OCR entirely by using machine-readable AA data instead. Offer this pathway in the Day 7 nudge as the "if upload is difficult" option.
The nudge sequence's effectiveness depends on the nudge content being genuinely helpful rather than just frequent. A nudge that says "your application is incomplete, please upload your documents" is less effective than one that says "you are missing your bank statement for the last 3 months — this can be downloaded from your bank's app in Settings > Account Statements." Content quality is a human curation responsibility that makes the difference between a nudge that recovers the application and one the borrower ignores.Review nudge open rate and document completion rate by template monthly — and rewrite any template with a completion rate below a configured floor. The best nudge content comes from understanding what specific friction the borrower is experiencing at each step, which the CX Strategy Officer AI's stall point data can inform.
The alternative document map is only as comprehensive as the credit policy allows. Where the credit policy has no approved alternative for a mandatory document — and some mandatory documents genuinely have no substitute — a borrower who cannot provide that document faces an impasse that no nudge sequence can resolve. This is a policy design issue, not a document ops issue.Review the mandatory-document-with-no-alternative list annually with the credit team — identifying whether any high-abandonment mandatory documents could have an approved alternative added to the policy without materially increasing credit risk. The alternative map's comprehensiveness is a credit policy decision that directly affects conversion rates.
Agent Profile · Document Ops Agent AI · LendingIQ · BengaluruLast updated April 2026 · For internal use

Important Reads

Learn more about how to deploy Document Ops Agent AI to your lending workflow.