Merchant onboarding used to take 4 days. Now it takes 4 hours. Here is the breakdown of how we achieved this for a mid-sized private sector bank.
The Bottleneck: Name Matching Hell
The bank’s operations team was spending 80% of their time on one specific problem: Name Mismatches. The name on the PAN card was “M/S Enterprises Pvt Ltd”, but the GST certificate said “M/S Enterprises Private Limited”. Or “Kumar A.” vs “Anil Kumar”.
Traditional fuzzy matching algorithms yielded too many false positives, forcing humans to manually review every single case. The queue was endless.
The Solution: Intelligent Fuzzy Logic
We deployed an LLM-based agent specifically trained on Indian corporate entity naming conventions. The agent understands that “Pvt Ltd” and “Private Limited” are identical. It understands that “Gurgaon” and “Gurugram” are the same city. It understands that ” & ” and ” and ” are interchangeable.
The Workflow
- Step 1: OCR extracts data from all uploaded docs.
- Step 2: The Agent compares the extracted entities.
- Step 3 (Auto-Approve): If the match confidence is >98%, the system auto-approves. (This cleared 60% of the queue instantly).
- Step 4 (Augmented Review): For the remaining 40%, the agent generates a “Reasoning Summary”. e.g., “Partial Match: Names match, but address differs slightly (Block A vs Block B). Risk Level: Low.”
The Result
The ops team went from reviewing 500 cases a day to handling complex exceptions only. The backlog vanished in week one. More importantly, the merchant drop-off rate reduced by 25% because they were getting their QR codes active the same day.