The job before a coworker owned it
ABFL underwrites thousands of retail and small-business loans every month, and every one of them used to start the same way: an analyst manually assembling a Credit Appraisal Memo from scratch.
Bank statements, GST filings, payroll slips, and financial ratios were pulled across multiple tools by hand.
Each CAM took 2 to 3 hours of copy-paste, manual calculation, and analyst coordination.
Compliance evidence and audit trails were not captured systematically as the work happened, only reconstructed later, under pressure, when a regulator asked.
Turnaround scaled directly with loan volume, which meant growth made the bottleneck worse, not better.
What the coworker owns end to end
The Underwriting coworker collapses document ingestion, extraction, and CAM generation into one continuous case instead of a chain of manual handoffs.
Ingests bank statements, GST data, and payroll slips automatically within seconds of a document landing, no manual curation required.
Extracts over 1,800 structured fields per borrower through OCR and NLP, normalized into the financial variables a CAM actually needs.
Auto-generates the CAM itself, embedding explainable risk scores like PD and LGD and flagging policy breaches directly, rather than leaving that judgment to be reconstructed manually.
Surfaces everything inside the existing Loan Origination System, so underwriters see the case, not a separate tool they have to switch into.
Hands off to an analyst only for exceptions, the cases that actually need a person's read, not the ones that were always going to be a formality.
Three document sources are ingested and normalized into one audit-ready CAM.
The numbers
70% reduction in underwriting cycle time overall.
CAM generation dropped from 2 to 3 hours to under 40 minutes per case.
16,000+ financial documents parsed every month, up from roughly 3,500 before.
Audit readiness compressed from 10 days to 2 days per quarter.
Customer-facing turnaround improved from 4 days to 6 hours.
First-cycle delinquency improved by roughly 18 basis points, and borrower NPS rose by about 14%.
In their words
"Integrating Tartan's unified APIs took less than a sprint and eliminated weeks of manual spreadsheet work."— Rajesh Shetty, Chief Risk Officer, ABFL |
"Our analysts focus on exceptions now. The AI handles the grunt work and produces an audit-ready CAM every time."— Divya Narain, VP Credit Ops, ABFL |
What this means for the credit risk team
A credit analyst's judgment was always the scarce resource. Two to three hours per CAM meant most of that judgment was going into data assembly, not risk assessment. The Underwriting coworker owns the assembly end to end and hands the analyst exactly the cases where a person's read actually changes the outcome, which is the only place that judgment was ever meant to go.
