Case Study · Data Platform · Microsoft Fabric

Building an AI-ready data foundation on Microsoft Fabric for a mid-market enterprise

A professional services firm with data scattered across 14 disconnected systems needed a single, governed data foundation before any AI ambitions could become real. We built it in 16 weeks.

Sector Mid-Market Enterprise

Engagement Project Delivery

Timeline 16 weeks

Stack Microsoft Fabric · dbt · Power BI

The Challenge

AI ambitions on a broken data estate

The client had already spent 18 months and significant budget attempting to build AI-powered dashboards and forecasting tools. Every project stalled for the same reason: data quality. Fourteen source systems, no common definitions, no single source of truth — and a management team that had lost confidence in any number a data system produced.

The Problem

Fragmented data made every AI project a false start

14 disconnected systems — CRM, ERP, billing, project tools, spreadsheets

No agreed data definitions — 4 different revenue figures in 4 different systems

Monthly reporting cycle — manual, error-prone, 4 days to produce

No data lineage or governance framework in place

Two failed AI projects — both blocked by data quality issues

Our Approach

Foundation first — then the AI on top of it

AI-Ready Blueprint™ — full data audit and architecture design in 3 weeks

Microsoft Fabric lakehouse — single unified data store with medallion architecture

dbt transformation layer — documented, tested, version-controlled

Unified semantic layer — agreed definitions, single source of truth

Power BI executive dashboards — live, not monthly

Full knowledge transfer — in-house team owns and operates from day one

Platform Architecture

Medallion architecture on Microsoft Fabric

A layered, medallion-style architecture that separates raw ingestion from curated, governed data — giving the business both operational confidence and AI-readiness from the same platform.

Source Layer

14 Source Systems

CRM (Salesforce)ERP (SAP)Billing SystemProject ToolsExcel / SharePoint+ 9 others

Bronze Layer

Raw Ingestion — OneLake / Fabric Data Factory

Incremental loadsSchema-on-readAudit timestampsFull lineage

Silver Layer

Cleaned & Conformed — dbt + Fabric Notebooks

Data quality rulesdbt testsUnified entity modelAgreed definitions

Gold Layer

Business-Ready — Semantic Models & Feature Store

Semantic layerKPI definitionsAI feature storeRow-level security

Consumption

Power BI · AI/ML workloads · Self-serve analytics

Executive dashboardsOperational reportingAI-ready endpoints

"We had tried twice to build AI on top of our data. Both times it failed. Marzal Labs told us the same thing both times had failed for the same reason — and fixed the actual problem before building anything else."

Chief Operating Officer · Professional Services Firm (name withheld)

Delivery Timeline

Foundation to production in 16 weeks

Weeks 1–3 · AI-Ready Blueprint™

Full data audit and architecture design

Mapped all 14 source systems, catalogued every data asset, identified quality issues, inconsistencies, and gaps. Produced the architectural blueprint for the Fabric lakehouse and a prioritised implementation roadmap with cost and timeline estimates.

Weeks 4–8 · Bronze + Silver

Ingestion pipeline and data quality framework

Built the Fabric Data Factory pipelines to ingest all 14 source systems into OneLake. Developed the dbt transformation layer with 180+ data quality tests. Established the unified entity definitions — the first time the business had an agreed definition of "customer" and "revenue".

Weeks 9–13 · Gold + Consumption

Semantic layer, dashboards, and AI endpoints

Built the gold-layer semantic models with full row-level security. Delivered 6 executive Power BI dashboards with live refresh. Created the AI feature store endpoints to support the client's next phase of ML workloads.

Weeks 14–16 · Handover

Documentation, training, and full ownership transfer

3 days of hands-on training for the in-house data team. Full dbt project documentation, Fabric workspace runbooks, and an operations guide. The client's data team now manages and extends the platform independently.

Results

The foundation that made everything else possible

14 → 1

Source systems unified into one governed lakehouse

4 days → 2hrs

Monthly reporting cycle, now running weekly with live dashboards

180+

Data quality tests running on every pipeline refresh

100%

In-house team ownership — no ongoing dependency on Marzal Labs

3 weeks

Time from engagement start to agreed data definitions across the business

AI-ready

Platform now supports the client's first ML forecasting workloads — live Q3 2026

Technology

Microsoft-native, open-standard, zero lock-in

🏗️

Microsoft Fabric

Unified analytics platform

🗄️

OneLake

Unified data storage

🔄

Data Factory

Ingestion pipelines

📐

dbt (Fabric)

Transformation layer

📊

Power BI

Dashboards & reporting

🔗

Fabric Notebooks

Python / Spark workloads

🔐

Entra ID

Identity & access

📋

Purview

Data governance & lineage