The challenge
Leadership wanted AI assistants grounded in firm knowledge, but engagement data carried strict confidentiality obligations. Access was inherited from years of project workspaces. Risk and legal teams would not approve Copilot without provable controls.
What we did
Tulan designed a domain model aligned to service lines, then used Loom Steward to scan and classify 2.3 million files and 14,000 tables. Sensitivity labels, DLP policies, and workspace roles were rebuilt on least privilege. Certified semantic models became the grounding layer for Fabric data agents, with verified answers for the most common partner questions.
The outcome
Risk and legal approved a phased Copilot rollout to 40,000 professionals. An external audit found no sensitive-data exposure. Partners now ask plain-language questions about pipeline, utilization, and realization and get answers grounded in certified data.
“Governance was the thing standing between us and AI. Tulan turned it into the thing that made AI possible.”
Chief Data Officer