Banking
Loan Processing & Document Collection Automation
A leading regional commercial bank faced significant delays in retail and SME loan processing due to fragmented document collection and manual verification workflows.
Loan officers spent hours chasing applicants for missing documents such as bank statements, GST filings, income proofs, and KYC records. Customers frequently uploaded incomplete or incorrect documents, forcing multiple back-and-forth interactions across email, WhatsApp, and phone calls. As a result, loan approval turnaround time often exceeded 7–10 days, and relationship managers had limited visibility into where applications were stuck.
Subverse Agent Flow
When a new loan application is submitted in the bank's loan origination system, Subverse automatically triggers a multi-agent workflow.
Agents involved:
Application Intake Agent
Pulls applicant details from the Loan Origination System
Creates an entity memory record for the borrower
Determines required documents based on loan type (personal, SME, mortgage)
Document Collection Agent
Sends structured document requests via WhatsApp, email, or voice
Accepts PDFs, images, and scanned documents
Validates file types and completeness before submission
Document Intelligence Agent
Extracts fields from uploaded documents (PAN, GSTIN, income data)
Flags discrepancies against application data
Routes incomplete or suspicious documents for human verification
Credit Coordination Agent
Pushes verified data to underwriting systems
Updates loan officers with real-time status summaries
If documents remain missing after defined intervals, the agent proactively schedules reminders or voice calls, escalating to relationship managers only when necessary.
Integrations
Loan Origination System (LOS)
Core Banking System
Document Management System
OCR / document extraction APIs
WhatsApp Business / Email / Voice telephony
CRM for relationship manager notifications
Measurable Outcome
Loan processing TAT reduced by 55–65%
80% of document collection handled autonomously
Relationship manager follow-ups reduced by 70%
Multimodal + Memory
Accepts documents, images, or voice uploads
Maintains borrower-level memory across application stages
Tracks document history and missing items across channels
