Skip to main content
SupaCortex
Sign in Book a Demo
Industry

Answers that cite the approved document - or say there isn't one

In financial services, a confident wrong answer about a charge is worse than no answer. SupaCortex answers from the documents you have approved, shows where each sentence came from, and admits the gaps.

Book a demo See pricing
  • Official sources rank first
  • Says "not enough evidence"
  • A person sends every reply

How do banks, NBFCs and insurers use SupaCortex?

Financial services teams load approved product sheets, schedules of charges, policy wordings and process notes into projects marked as official sources, and keep internal guidance in separate projects for staff. Customers get cited answers on the website; agents get cited reply drafts that a person edits and sends; branch and operations staff ask internal procedures in team chat. When the approved material does not answer a question, the assistant says so.

  • Trust tiers: official documents outrank standard material and past tickets.
  • A confidence gate returns "insufficient evidence" instead of guessing.
  • Every answer carries checked, numbered citations.
  • The AI never sends a reply - an agent does.
  • Open models on infrastructure we operate; no third-party AI API.
Typical questions

Questions it answers in financial services

"What's the foreclosure charge on a personal loan?"

Quoted from the current schedule of charges, with the clause cited.

"Which documents do I need for a home loan?"

From the approved checklist - not from what an agent once wrote in a ticket.

"Is dental covered under my policy?"

From the policy wording, and "not enough evidence" if the wording is silent.

"When is my next EMI due?"

In identified mode, a lookup on your loan system for the signed-in customer only.

"What's the KYC re-verification process?"

Asked by branch staff in team chat, answered from the internal operations manual.

A complaint about a deducted charge

Drafted for an agent with the relevant policy cited; the agent decides what goes out.

Set-up

How financial services teams set it up

  1. 01

    Public and internal apart

    A customer project with approved material only; separate projects for operations and branch staff.

  2. 02

    Mark the official sources

    Product sheets and charges as trust tier 1, so they always outrank anything else.

  3. 03

    Widget on the website

    Public mode for product questions; identified mode in net banking or the app for account questions.

  4. 04

    Drafts, not auto-replies

    Agents start from cited drafts in the help desk and keep the final word.

Honest limits

What to know before you start

We would rather your compliance team hears these from us.

  • No SOC 2 or ISO 27001 certification today, and no choice of data-residency region.
  • It does not give financial advice or make credit decisions - it quotes your documents.
  • Access is by project, not by mirroring per-user permissions from SharePoint or Drive.
  • No WhatsApp channel yet; it is on the roadmap.

Read next

Answers With Citations

Hybrid search, a confidence gate and checked citations - it says "I don't know" instead of guessing.

Security

Separate databases per workspace, encrypted secrets, and answers scoped to who is asking.

Reply Drafts for Help Desks & Email

Grounded, cited reply drafts inside UnifyDesk or any desk on the Draft API. A person still sends.

Email Reply Assistant

Draft answers to customer and partner emails from your knowledge, in the help desk you already use.

Roles, SSO & Two-Step Sign-In

Owner, Admin, Editor and Viewer roles, Google and Microsoft sign-in, and authenticator codes.

Measuring AI Answer Quality

Golden question sets, groundedness, deflection and the numbers that actually predict trust.
Frequently asked

Questions people ask before they start

Mark approved product documents as trust tier 1 (official). Official documents outrank standard and past-ticket material, and when enough official evidence exists past-ticket passages are dropped. Keep anything unapproved out of customer-facing projects altogether.

It answers only from retrieved passages, cites each one, and its citations are checked before the answer is shown. When nothing relevant is found it says there is not enough evidence. No system is perfect, which is why each answer shows its source for the customer or agent to check.

No. Reply drafts go to an agent in the help desk, who edits and sends. In UnifyDesk, your approval rules still apply to that reply.

Not today, and we would rather say so than imply otherwise. Each workspace has its own databases, secrets are encrypted, and documents are not sent to a third-party AI API. We are happy to answer a security questionnaire.

In identified mode, yes - your backend signs a token for the logged-in customer and you describe the lookups allowed on your API. The customer identity comes from the token, never from the chat.

Ask it your trickiest charges question. Then check the citation.

Book a 30-minute demo on your own documents. We set up a 14-day trial workspace for you - no credit card, no setup fee.

Book a demo See pricing

Questions first? Email info@greatdevelopers.com or call +91 8585953085.

Still curious? Ask your favorite AI about SupaCortex.

ChatGPT Claude Perplexity Grok