A professional services firm in the mid-market
Fourteen disconnected systems
into one governed foundation
Sixteen weeks from legacy sprawl to a production AI-ready platform, owned by the in-house team from handover.
- Sector
- Mid-market enterprise
- Engagement
- Production deployment
- Timeline
- 16 weeks
- Stack
- Microsoft Fabric · dbt · Power BI
The challenge
What the team was living with
The client had already spent eighteen months and a significant budget attempting to build AI-powered dashboards and forecasting tools. Every project stalled for the same reason, and it was never the model: fourteen source systems, no common definitions, and no single source of truth anywhere in the estate.
Four systems produced four different revenue figures. The consequence was predictable — the management team had stopped trusting any number a data system produced, which made every subsequent data investment harder to justify than the last one.
The monthly reporting cycle took four days of manual assembly and was out of date on the day it landed.
- 14 disconnected systems — CRM, ERP, billing, project tools, spreadsheets
- 4 different revenue figures in four different systems
- Monthly reporting cycle — manual, error-prone, four days to produce
- No data lineage or governance framework anywhere in the estate
- Two failed AI projects, both blocked by data quality rather than technology
- Management confidence in reported numbers effectively at zero
The approach
Foundation first.
Then the AI on top of it
Audit the estate before designing anything
A three-week AI-Ready Blueprint catalogued every source system, its quality profile and the competing definitions in use. The audit shaped the architecture; architecture drawn before the audit would have been architecture for a different company.
Build the medallion lakehouse
A Microsoft Fabric lakehouse with Bronze raw ingestion into OneLake, Silver cleaned and conformed, and Gold business-ready semantic models. Each layer purpose-built, documented and independently testable.
Put the transformation layer under version control
dbt for all transformations — documented, tested and reviewable like any other code. This is what made the 180+ quality tests possible and the logic portable.
Agree the definitions
The most transformative work on the engagement, and the least technical: one agreed definition each for revenue, customer, project and cost, documented and respected by every downstream system. This is what ended the four-revenue-numbers problem.
Replace the monthly cycle with live dashboards
Power BI executive dashboards built on the Gold layer, refreshing weekly rather than assembled manually each month — and every figure traceable to source through lineage.
Hand over completely
Full dbt project documentation, Fabric workspace runbooks, pipeline operations guides and hands-on training. The in-house team has owned and operated the platform since handover.
The architecture
Medallion architecture on Microsoft Fabric
A layered architecture that separates raw ingestion from curated, governed data — giving the business operational confidence and AI-readiness from the same platform rather than from two competing ones.
- CRM
- ERP
- Billing
- Project tools
- Excel / SharePoint
- + 9 others
- OneLake raw ingestion
- Incremental loads
- Audit timestamps
- Full lineage
- dbt transformations
- 180+ quality tests
- Unified entities
- Conformed schema
- Business-ready semantic models
- Agreed definitions
- Documented business logic
- Power BI executive dashboards
- ML endpoints
- AI-ready feature layer
Outcomes
What changed, measured
The AI ambitions that had failed twice became straightforward once the foundation existed, which was always the point. The change the management team notices day to day is smaller and more important: when two people quote a revenue figure, it is now the same figure.
Take it with you
The full engagement, in twelve pages
Architecture decisions, what we got wrong on the way, and the measurement methodology behind every figure on this page.