Data Platform Architecture — WYCA Learner Destinations

← Back to Destinations Index → Tender Submission (Q2.3)

Architecture (as submitted)

Technical architecture: identifiable data in IBP's Snap platform; a Microsoft Fabric pipeline pseudonymises via the Snap API and loads a Lakehouse/Azure SQL analysis database behind a Power BI dashboard; two hosting options — From learner response to live dashboard — identifiable data never leaves the secure survey platform. Open full size ↗

Proposed hosting design

The Fabric pay-as-you-go F SKU can be paused when not in use. The design splits the platform across two workspaces so reports stay available all day while compute is paused overnight.

ETL — Fabric F2 PAYG

ETL Fabric Workspace

Hosted on a Fabric F2 pay-as-you-go capacity. The capacity is automatically resumed before the nightly ETL and paused once processing completes.

Reporting — shared Power BI Pro

Reporting Workspace

Hosted on shared (Power BI Pro) capacity using an Import semantic model. Because the data is stored in the model, reports remain available throughout the day even while the F2 capacity is paused.

Keeping the reporting workspace on shared capacity ensures reports remain accessible without requiring the Fabric capacity to be running.

The analysis database will be a Fabric Lakehouse or Warehouse (the submitted architecture figure labels it "Lakehouse / Azure SQL").

Azure resources required

Azure Pay-As-You-Go subscription
Resource Group
Fabric F2 Capacity
Azure Automation Account (to schedule capacity pause / resume)
Existing Microsoft Entra tenant

Indicative monthly costs

Item	Indicative cost
Data platform
Fabric F2 capacity (1 hour/night)	£8–10 / month
Azure Automation	~£1 / month
OneLake storage	~£1 / month
Estimated total Azure / Fabric cost	£10–12 / month
Power BI shared capacity
Power BI Pro licence (users may already have these)	£12–14 / user / month

Why pause/resume: leaving the F2 capacity running 24/7 would cost roughly £200–220 per month. Automating pause/resume so the capacity only runs for the nightly refresh brings the platform cost down to ~£10–12 per month — a significant saving, and the same cost discipline that would apply at 45,000-learner scale.

Source: proposal from Matthew Denison (Data Platform Lead) to Chris Maunder. Aligns with the technical architecture in the final submission (Q2.3).