Power BI
Microsoft Fabric
2025
8 min read
Power BI Premium to Microsoft Fabric: Investment Management Migration Case Study
Sector
Asset Management
AUM
approx. GBP 18 billion
Engagement
22 Weeks · 300+ BI Users
Offices
London (HQ) · New York
Numlytics delivered a complete Power BI Premium to Microsoft Fabric migration for a UK-listed investment management firm with approximately £18 billion AUM , migrating 214 reports from a Premium P3 estate to Microsoft Fabric F64 with zero downtime, reducing BI licensing costs by 55%, cutting report refresh from 45 minutes to under 90 seconds via Direct Lake, and satisfying FCA data lineage requirements for all 67 regulated reports through Microsoft Purview.
The Challenge: A Premium P3 Estate Growing Faster Than the Firm Could Govern It
The firm had built its Power BI Premium capability seriously over five years - growing from two analysts to twelve, and expanding the report estate to cover Risk, Portfolio Management, Client Reporting, Compliance, and Finance. By 2024, they were running 214 reports off a Premium P3 licence at a cost that had nearly tripled since the original commercial agreement.
-
Data staleness during trading hours: Most datasets ran in import mode, meaning portfolio managers and risk analysts were frequently looking at data that was between one and four hours old during market hours. Aggressive refresh schedules were blocked by Premium capacity throttling at peak load
-
FCA governance exposure: A review triggered by an FCA supervisory visit found that data lineage could not be demonstrated from the source system to the dashboard for 67 of their regulated reports, a significant regulatory exposure outstanding for over a year
-
Eleven conflicting measure definitions: Risk-adjusted return calculations had at least two different definitions in active use across different teams, producing results that could diverge by up to 8% depending on which report a stakeholder opened
-
Escalating licence costs: Premium P3 costs had nearly tripled, while Copilot ambition and Fabric Data Engineering workloads were unavailable at P3 - making the commercial case for migration unavoidable.
Eleven measures calculating risk-adjusted returns had at least two different definitions in active use. Results could diverge by up to 8% depending on which report a portfolio manager opened. This was not a technology problem. It was a governance problem that technology had made invisible, and it had already triggered an internal audit finding.
The Numlytics Approach:
Numlytics structured this Power BI Premium to Microsoft Fabric engagement around one non-negotiable constraint: 300 active users in a live trading environment could not experience any disruption to reporting. Every technical decision was made in service of that constraint.
-
Discovery and Governance Remediation (Weeks 1–5)
The Power BI Metadata Scanner API gave a complete inventory of all 214 reports, 89 datasets, and 67 data sources. Numlytics identified the 11 conflicting measure definitions as the highest-priority remediation item and ran two workshops with Risk and Portfolio Management to agree on canonical definitions before any migration work began, involving the Head of Data Governance to resolve genuine disagreements about how certain adjustments should be treated. 31 reports with no active users in the past 90 days were identified for retirement.
-
Fabric F64 Architecture and Pilot Migration (Weeks 6–10)
The target architecture was designed: Microsoft Fabric F64 with domain-aligned workspaces (Risk, Portfolio, Compliance, Finance, Client Reporting), a consolidated Lakehouse on OneLake with Bronze, Silver, and Gold layers in Delta format, and Direct Lake semantic models replacing all import-mode datasets. A pilot cohort of 22 reports across Risk and Portfolio Management was migrated first, running in parallel with the existing Premium environment for two weeks to allow side-by-side validation before any cutover decision.
-
Bulk Migration and FCA Regulated Reporting (Weeks 11–18)
With the pilot validated, the remaining 183 reports were migrated in four waves, sequenced to avoid month-end and quarter-end reporting periods. All 15 ADF pipelines were migrated to Fabric Data Factory with improved monitoring and audit logging. All 67 FCA-regulated reports had full data lineage documented in Microsoft Purview before the Premium licence was switched off, resolving a regulatory exposure that had been outstanding for over a year.
-
Copilot Enablement and Semantic Model Certification (Weeks 18–22)
A certified semantic model framework was established: only models that had passed data quality validation, lineage review, and compliance sign-off could be endorsed for use with Copilot. Copilot in Fabric was enabled for all 300+ users, with access to endorsed models only. A self-service analytics portal replaced the previous ad-hoc request process in the final week.
Before Numlytics
-
45-minute import-mode refresh cycles during trading hours
-
11 conflicting measure definitions, results diverged up to 8%
-
Zero FCA data lineage coverage for 67 regulated reports
-
Premium P3 costs nearly tripled from original agreement
-
No self-service analytics, ad-hoc request backlog growing
After Numlytics
-
Under 90 seconds via Direct Lake, live data during trading hours
-
Single certified measure library, consistent results across all reports
-
Full Purview lineage for all 67 regulated reports - FCA exposure closed
-
Fabric F64 at 55% lower cost than Premium P3
-
300+ users on Copilot self-service analytics portal
The Results
55%
Licence Cost Reduction
Power BI Premium P3 to Microsoft Fabric F64 - consolidated capacity
90 sec
Direct Lake Refresh
Was 45 minutes import mode - portfolio managers on live data during trading
100%
FCA Lineage
All 67 regulated reports with full Purview lineage - regulatory exposure closed
Zero
Downtime
All 214 reports live in Fabric - not a single user raised a support ticket
300+
Copilot Users
Live within 5 weeks of cutover on certified, governance-approved models.
"We had been postponing this migration for eighteen months. The reason was simple: we were running 214 reports in a live trading environment and could not afford disruption. Numlytics gave us the confidence to proceed by running a rigorous parallel validation before any cutover. When we finally switched off Premium, not a single user raised a support ticket. The Direct Lake performance improvement was immediate and dramatic. We now have portfolio managers checking live data on Copilot during the trading day, which simply was not possible before."
— Head of Technology and Data, UK Investment Management Firm, London