How a UK Specialist Bank Tested AI-Powered Mortgage Underwriting on Copilot Studio
One Orchestrator and seven specialist agents on Microsoft Copilot Studio take on the reading, cross-checking and arithmetic of a mortgage…
Read the case studyA UK digital bank's users ask one Copilot assistant in Teams and Dynamics 365 and get governed answers from Snowflake, SharePoint and an external API, under their own permissions.
THE CHALLENGE
The bank’s teams work in Teams, Outlook and Dynamics 365. The answers they need sit in Snowflake, SharePoint and third-party systems. Every question that crossed those systems meant a dashboard change, an analyst ticket, or someone stitching exports together by hand.
Generic AI assistants made trust worse: each invented its own definition of revenue or margin. Shortcuts such as exporting data into Microsoft 365 or sharing one service account bypassed the row-level and masking policies that protect customer data.
Before
THE APPROACH
We set up Copilot Studio as a multi-tool router. It reads each question, picks the tools it needs, and merges the partial answers into one reply. Numbers come from a Snowflake Cortex Agent, documents from a SharePoint site, and live third-party data from an external API.
We assessed four routes to Snowflake and chose MCP with one governed agent: a standard protocol, per-user identity, and one governed Snowflake tool instead of raw query tools, so all data reasoning stays inside Snowflake.
Question in Teams or D365
Business users ask in plain language, where they already work
Copilot Studio router
Sends each part to Snowflake, SharePoint or the external API
One governed answer
Merged reply, each source under its own access rules
THE OUTCOME
Users now ask in Teams or Dynamics 365 and get one answer from 3 sources, with no dashboard request or analyst ticket. Every metric has 1 definition, set in Snowflake semantic views. Snowflake queries run as the signed-in user, only the question and the final answer cross into Microsoft 365, and there’s no middleware to host or patch.
Outcome
- 3 sources answered in one reply
- 1 governed definition per metric
- Snowflake queries run as the signed-in user
- 0 middleware to host or patch
Copilot sees one governed Snowflake tool, not raw query tools, so all data reasoning stays inside Snowflake.
Put governed answers in front of your business users. IWConnect designs and delivers Snowflake AI solutions end to end, from the semantic layer and the knowledge graph to Copilot. Talk to us to plan a pilot.
iwconnect.com | sales@iwconnect.com
Tell us what you are trying to connect or trust. We will show you the proof that fits and the team that can help.
Thank you! Your Case Study is ready.
Get it Now!By signing up for the waiting list now, you'll secure your spot for early access and claim these valuable benefits.