For the last two decades, the corporate intranet has been the place where documents go to die. We’ve all seen it: a sprawling SharePoint site or a custom portal that requires five clicks to find the leave policy, ten clicks to find the latest credit risk circular, and a master’s degree in library science to locate the SOP for a specific trade finance product.
The Problem with Search in Banking
Traditional keyword search fails in banking because banking queries are inherently complex and contextual. A relationship manager (RM) doesn’t just want to “find a document.” They have a specific scenario in mind.
Consider this query: “Can I offer a 50bps waiver to a Tier-1 customer with a CIBIL score of 780 for a home loan?”
A keyword search engine will return:
- The 2024 Home Loan Policy PDF (50 pages)
- The 2023 Pricing Circular (deprecated)
- A generic FAQ on interest rates
The RM then has to download the PDF, CTRL+F for “waiver”, check if “Tier-1” definitions have changed, cross-reference with the latest risk circular on CIBIL cutoffs, and hope they haven’t missed an amendment released yesterday.
GenAI changes this completely. It acts as a reasoning engine, not just a retrieval engine. It ingests the policy, the circulars, and the amendments, and returns:
“Yes, per the Q3 2025 Pricing Circular (Section 4.2), Tier-1 customers with CIBIL >750 are eligible for up to 75bps discretion. However, this requires Zonal Head approval as the standard branch discretion is capped at 25bps.”
Permission-Aware Retrieval (The Hard Part)
The biggest hurdle to AI adoption in banks has been security. You can’t just dump all your documents into a massive vector database and let everyone query it. A junior analyst shouldn’t see the CEO’s compensation details just because they asked, “What are the salary bands?”
This is where RAG with RBAC (Role-Based Access Control) comes in. In our architecture, the AI doesn’t just search the database. It first checks the user’s Active Directory entitlements.
1. User Query: “Show me the NPA report for Mumbai Zone.”
2. Identity Check: User is a Branch Manager in Andheri.
3. Filter: The vector search is restricted ONLY to documents tagged with view_permission: Mumbai_Zone or view_permission: Public.
4. Retrieval: The system retrieves the relevant report.
5. Generation: The AI summarizes the report.
If a Chennai Branch Manager asked the same question, the system would return “I cannot access data for Mumbai Zone.” This prevents the “Chinese Wall” breaches that compliance teams lose sleep over.
From Retrieval to Action: The Agentic Shift
The legacy intranet was read-only. The new AI knowledge layer is read-write-execute. We are moving from “Chat with your PDF” to “Chat with your Bank.”
Instead of just finding the “Leave Policy,” the AI can:
- Check Balance: “You have 12 days of privilege leave remaining.”
- Predict Impact: “Taking leave in the first week of April overlaps with the Quarter End audit closing. Approval might be rejected.”
- Execute: “Draft a leave application for Oct 24-28 citing ‘Personal Wedding’ and send it to my manager.”
This capability—to bind knowledge retrieval with API execution—is what transforms the intranet from a document graveyard into the bank’s central nervous system.
Conclusion
The era of searching through folders is over. Banks that deploy a conversational, agentic knowledge layer will see faster onboarding, fewer compliance errors, and a significant reduction in operational friction. The intranet is dead. Long live the AI Brain.