← Agent catalogue
AI Agent Profile · LendingIQ · Bengaluru
Loan Origination Agent AI
Function: Loan Processing — Application StageRuntime: AWS Bedrock · ap-south-1Model: Claude Sonnet 4Context window: 200K tokens
DivisionOnboarding
Resume
What this agent does
The Loan Origination Agent AI receives every new loan application, checks it for completeness against the product-specific document checklist, enters the verified application data into the Loan Origination System, triggers the bureau pull, and assembles the complete application file for handoff to the Credit Underwriting Agent AI. It is the intake and preparation function of the credit pipeline — ensuring that every application that reaches underwriting is complete, accurately entered, and bureau-enriched. It processes; the Underwriting Agent decides.
Primary functions
Application Intake & Doc Checklist
Per application — on submissionInvoked when: application is submitted through any origination channel (direct digital, DSA, branch)
- Reads the submitted application data and document set against the product-specific document checklist retrieved from the policy RAG: which documents are mandatory for this product type, customer category (salaried / self-employed / MSME), and loan amount band. Produces a completeness verdict — all mandatory documents present, or a specific list of missing items.
- Checks basic eligibility pre-conditions before investing bureau pull cost: applicant age within the product's age band, residential address in a serviceable geography, loan amount within the product's range. Hard mismatches on these pre-conditions are returned immediately with the specific reason — no bureau pull is triggered for clearly ineligible applications.
- Sends a structured document request to the applicant (via the channel configured for this origination type) listing the specific missing documents with clear descriptions of what is required for each. The request uses approved templates — the agent does not compose novel communication.
Output: Completeness verdict — all documents present or specific missing items listed with document descriptions. Eligibility pre-check result. Document request sent to applicant if incomplete.
LOS Data Entry
On complete application — before bureau pullInvoked when: application completeness check passes
- Extracts all structured data fields from the submitted application and documents — applicant name, date of birth, address, PAN, income declared, employer/business name, loan amount requested, tenure requested, purpose code — and enters them into the Loan Origination System with field-level data quality checks. Flags any field where the extracted value conflicts across documents (name spelling on PAN vs Aadhaar) for human processing officer review before the LOS record is finalised.
- Creates the LOS application record with a unique application reference, links all submitted documents to the record in the document management system, and records the origination channel, origination date, and the DSA/branch code where applicable for commission and attribution tracking.
Output: Populated LOS record with all extracted fields, document links, origination metadata, and field conflict flags for human review.
Bureau Pull
On LOS record creation — before underwriting handoffInvoked when: LOS record is created and all mandatory documents are linked
- Triggers the bureau pull for the applicant (and co-applicant/guarantor where applicable) using the PAN as the primary identifier. Selects the bureau based on the product configuration — CIBIL for most products, with Experian or CRIF as configured for specific product types or as a secondary pull for NTC borrowers.
- Appends the bureau report to the LOS application file and records the enquiry in the application audit trail. If the bureau pull fails (API timeout, PAN not found, system unavailability), logs the failure, retries once after 30 seconds, and flags to the human processing officer for manual bureau pull if the retry also fails.
- Does not interpret the bureau report — that is the Underwriting Agent AI's function. The bureau report is appended to the file and passed forward. The origination agent's bureau step is a data assembly step, not an analysis step.
Output: Bureau report appended to LOS file, bureau pull status logged (success/failure/retry), and complete application file marked ready for handoff to Credit Underwriting Agent AI.
Hard guardrails
Will notAssess creditworthiness or apply the credit policy. The intake agent processes the file; the Underwriting Agent decides on the credit. These are structurally separated functions.
Will notTrigger a bureau pull for an incomplete or ineligible application. Bureau pulls have a cost and create enquiry records on the applicant's bureau file — they are not triggered until the application passes completeness and basic eligibility checks.
Will notOverride a missing mandatory document requirement. Missing documents result in the application being returned to the applicant — the intake agent does not waive mandatory document requirements.
Known limitations
Document completeness checking is against the configured checklist — if the product policy changes and the RAG checklist is not updated within 24 hours, applications will be checked against the prior checklist. A policy change that adds a new mandatory document will not be enforced until the checklist is updated.Build a checklist version control check into the LOS data entry step — the checklist version applied to each application is logged in the LOS record. If the policy team updates the checklist, applications already in process under the prior version are reviewed and supplemented as needed.
OCR extraction of unstructured documents (hand-written income certificates, non-standard salary slip formats) may produce extraction errors that require human review. The agent flags extraction confidence levels below a threshold and sends those fields for human processing officer verification — but the flagging relies on the OCR engine's confidence scoring, which may not catch all errors.Implement a periodic QC sampling process: a human processing officer reviews a random 5% sample of LOS records for OCR accuracy, particularly for document types with high error rates. QC findings feed back into the OCR engine's training data for that document type.
Important Reads
Learn more about how to deploy Loan Origination Agent AI to your lending workflow.
- Use case #0001From application to credit queue in 4 minutes: how Origination AI worksA loan application that takes four minutes to process from submission to credit queue is not a streamlined manual process — it is an automated one. The Loan Origination Agent AI extracts every field from the application and its documents, runs identity and eligibility checks, pulls the bureau report, computes the preliminary underwriting metrics, and creates the LOS record — before a human underwriter has opened their laptop.Read article →
- Use case #0002Bureau pull automation: how Origination AI triggers and interprets bureau dataPulling a bureau report takes 20 seconds. Interpreting it — extracting the relevant metrics, identifying the material tradelines, computing the FOIR impact of existing obligations, flagging the DPD history, and translating the raw data into the three numbers the credit underwriter actually needs — takes 15 minutes when done manually and introduces errors when done under time pressure. The Origination AI interprets the bureau report in the same 20 seconds it takes to receive it.Read article →
- Use case #0003LOS integration: how Loan Origination AI connects to your core lending systemThe Loan Origination Agent AI does not replace a Loan Origination System — it populates one. The AI handles the work that happens before the LOS record exists: document extraction, field validation, bureau computation, and eligibility screening. The moment that work is complete, the AI writes the result into the institution's existing LOS through a structured API integration that requires no change to the LOS itself and no data migration from the institution's existing system of record.Read article →