Case Study
At a glance
The data landscape
Profile
Global asset manager
More than a dozen business units
Investment, ESG, research and market reference data
Environment
Snowflake on Azure
Microsoft 365: SharePoint and OneDrive
Microsoft Purview
File shares
The challenge
No way to tell which data was sensitive
No full picture of the estate. The firm could not say everything that existed across its data, or who could reach it.
Labels that said nothing. A default label marked nearly every document confidential, so labels could not separate sensitive data from the rest.
AI held back. AI teams were fed only data someone had checked by hand. Everything else stayed closed.
“We don't have our arms around everything that exists within the estate.”
A data leader at the firm
“We have a default label of confidential that hits everything.”
The firm's Head of Information and Records Governance
Why Bedrock Data
What the proof of value showed
In-environment scanning
Scanners ran inside the firm's own Azure environment across a Snowflake account and two SharePoint sites, in about 27 hours.
The firm's own taxonomy
Snowflake tables were sorted into the firm's own data domains, including a custom ESG category.
Entitlement mapping
Users, groups, service accounts and external shares, tied to the data they can reach.
Lineage by similarity
Data similarity traced where the same data appeared in more than one place.
Labels from findings
Bedrock findings will drive Purview sensitivity labels in place of the default.
Masking in Snowflake
Tags on Snowflake objects will drive masking, so policy follows the data.
Results
Results from the proof of value
The full scope scanned in about 27 hours. Snowflake and SharePoint, inside the firm's own cloud.
84% of 514 tables mapped automatically. Snowflake tables were assigned to the firm's own data domains.
Access mapped end to end. The team could see which identities reach each data set.
A three-year agreement. The firm signed in September 2026 and is rolling out to more than 2 PB of cloud data and 200 TB of file shares.
The proof of value was deliberately scoped and used low-sensitivity data. The firm chose to roll out on that basis and is now deploying into production.
The vision
Opening data to AI, minus the sensitive 1%
“At the moment, you can't have access to any of it. In the future, once this is fully operational and it's gatekeeping, you can have all of it and we'll just be blocked from that 1%.”
A data leader at the firm
What's next
Where this is heading
Snowflake and Microsoft 365 come first, then Purview labels and Snowflake masking. About 40 custom data types, mostly national identifiers, are being built for the firm's global footprint.
