Power BI
Microsoft Fabric
2025
8 min read
Microsoft Fabric Migration for Non-Profit Power BI Estate: 38% Cost Reduction Case Study
Sector
International Development
Location
London, UK
Engagement
20 Weeks
Platform
Microsoft Fabric F48 · Power BI
Numlytics delivered a complete Microsoft Fabric migration for non-profit Power BI estate for a UK-based Global Development Foundation - auditing all 173 reports, retiring 68 unused ones, migrating 105 active reports to Microsoft Fabric in two structured waves, and enabling Copilot in Fabric for 85 users within six weeks of cutover. The result: 38% reduction in data infrastructure cost and overnight data latency reduced to 30 minutes across all reporting datasets.
The Challenge: A Trust Problem at the Centre of Everything
A data team of three, supporting 80 users across programmes, finance, and executive leadership, had built a reasonably mature Power BI estate on a Premium P1 licence with Azure Data Factory pipelines pulling from Salesforce, an ERP, SharePoint document libraries, and a portfolio management system. On paper, this looked solid. In practice, the foundation had a significant trust problem.
-
Conflicting numbers between teams: Finance ran grant expenditure reports from one set of ADF pipelines. The programmes team produced dashboards from Salesforce exports and manually refreshed Excel files. When leadership asked about committed versus disbursed spend, they routinely received two different answers - triggering reconciliation meetings that consumed hours every week
-
Governance deficit across 173 reports: Nobody had a clear picture of which reports were authoritative, which were experimental, and which had been abandoned. Dataset ownership was undocumented. Several reports were still pulling from Excel workbooks in personal OneDrive folders
-
Overnight batch latency: All data pipelines ran on overnight schedules, meaning country office teams in Sub-Saharan Africa and South Asia were working from data that was already a full day old by the time they arrived at their desks
-
Copilot ambition blocked: Leadership wanted to move to Microsoft 365 Copilot, but had paused because the data team could not confidently say the underlying data was clean enough to trust AI-generated responses.
Of 173 reports in the Power BI tenant, 68 had not been opened in more than six months. This single finding from the estate audit gave the foundation an immediate and significant reduction in Microsoft Fabric migration scope before any technical work had started.
The Numlytics Approach:
Numlytics designed this Microsoft Fabric migration for non-profit Power BI as a five-phase programme, beginning with a rigorous discovery phase to establish what was worth migrating, and ending with Copilot enablement on certified data. No architecture decisions were made until the business had agreed on a single version of the truth.
-
Discovery and Power BI Estate Audit (Weeks 1–4)
Numlytics used the Power BI REST API and Microsoft Fabric Metadata Scanner to pull a full inventory of every report, dataset, and data source in the tenant. Each of the 173 reports was scored against three criteria: usage frequency over the past 90 days, business criticality as assessed by department heads, and data source complexity. The result was a Fabric Readiness Score and a prioritised migration backlog. Two ADF pipelines built by a contractor who had left two years earlier were identified during this phase, with no documentation and no monitoring.
-
Business Alignment on Metric Definitions (Weeks 3–5)
Before designing any architecture, Numlytics ran structured workshops with finance, programmes, and data leadership to agree on metric definitions in writing. What counted as committed spend? When was a grant considered disbursed? How should foreign currency transactions be treated? Getting these answers agreed before building the semantic model was the decision that made the entire Microsoft Fabric migration work, these were governance problems that technology had been obscuring, not solving.
-
Microsoft Fabric Lakehouse Architecture - OneLake Medallion (Weeks 5–14)
Numlytics designed and built a Microsoft Fabric Lakehouse on OneLake using a Medallion structure. The Bronze layer handles raw ingest from Salesforce, ERP, SharePoint, and the portfolio system via Fabric Data Factory. The Silver layer applies the agreed reconciliation logic and data quality checks. The Gold layer contains reporting-ready aggregates for each directorate - all in Delta Parquet format for ACID compliance and time-travel capability. All 11 ADF pipelines were migrated to Fabric Data Factory, including the two undocumented pipelines, which were reverse-engineered and rebuilt with full monitoring.
-
Governance, Migration and Cutover (Weeks 14–18)
Row-level security was implemented across all semantic models, aligned to the foundation's directorate structure. Microsoft Purview was configured for complete data lineage - satisfying donor reporting requirements around data quality and auditability. The 105 active reports were migrated in two waves: priority dashboards used by leadership and finance first, then programme-level reporting. The 68 unused reports were retired rather than migrated, reducing the ongoing maintenance burden immediately.
-
Copilot Enablement and Team Handover (Weeks 18–20)
Copilot in Fabric was enabled for the executive and senior leadership group using only the certified Gold-layer semantic models as the authorised data source. Sensitivity labels were applied across all data. A two-day internal handover workshop equipped the data team to maintain, extend, and govern the platform without ongoing dependency on Numlytics.
Before Numlytics
-
Overnight data latency - country offices working on day-old data
-
Two different answers to the same question, reconciliation meetings weekly
-
173 reports with no clear ownership or governance documentation
-
Copilot ambition paused - trust in underlying data too low
-
Two undocumented ADF pipelines running with no monitoring
After Numlytics
-
30-minute data latency - country offices working from same-day data
-
Single certified semantic model, conflicting reporting structurally impossible
-
Full Microsoft Purview lineage across all 105 active reports
-
Copilot live for 85 users, certified semantic models only
-
All pipelines documented, monitored, and alerting in Fabric Data Factory
The Results
38%
Cost Reduction
Migrating from Power BI Premium P1 to Fabric F48, consolidating BI and pipeline costs.
105
Reports Migrated
Zero disruption to reporting cycles - 68 unused reports retired
30 min
Data Latency
Was overnight batch, all key datasets now refresh every 30 minutes.
85
Copilot Users
Live within 6 weeks of cutover, certified semantic models only
"We had been talking about fixing our data for three years. The problem was that every fix we tried addressed a symptom without touching the root cause. Numlytics came in, identified that the fundamental issue was misaligned metric definitions rather than bad technology, and built the governance framework before they built the platform. The result is that our leadership team can now ask Copilot a question about grant disbursements and trust the answer. That was not something we thought was possible twelve months ago."
— Senior Data Manager, Global Development Foundation, London