The Fabric Maturity Roadmap
What each of the four maturity stages actually means, and what being at one gets you.
Four stages, and they have to be climbed in order.
Most capability models are a menu — pick the bits that appeal. This one is a ladder, and the rungs are load-bearing. Business users cannot safely self-serve on models that disagree with each other, and an agent cannot give consistent answers over data whose definitions live in people's heads. Each stage is built on the one below it.
That has a practical consequence worth stating plainly: the lowest unmet rung is the one that matters. A team with a working AI pilot and an unreliable warehouse is not at Stage 4 — it is at whatever stage its warehouse supports, with an impressive demo sitting on top. Knowing which rung you are actually on is the point of the exercise, and it is usually one lower than it feels.
The four stages
Foundational to advanced, left to right.
Engineered data warehouse / lakehouse
- Medallion architecture — raw → cleaned → business-ready
- AI-enabled development — plugins carry your standards
- Alerting — automated retry and fix processes
- Process quality — data reliability and compute use
- Ready for AI consumption — by end users
BI and FP&A analyst self-service
- Semantic model — standard schemas, minimized sprawl
- AI-assisted dashboard builds — Power BI, Fabric Apps, Claude Artifacts
- Process automation build-outs — Azure Logic Apps
- Git integration and deployment pipelines — source control, promotion path
Business user self-service
- AI tooling with skills — on top of the warehouse
- Awareness and training
- Standard build procedures and definitions — plugin-enabled development
- Self-service publishing within guardrails — business users ship safely
AI-focused workstreams
- Fabric IQ — a business-level map your AI tools can query
- Model selection and planning
- Microsoft Copilot Studio — agents for end users
- Azure AI Foundry toolboxes — curated tool sets for pro-code work
What being at each stage actually means
The capability list above is what you build. This is what it buys you.
Engineered data warehouse / lakehouse
Data arrives reliably, problems announce themselves, and someone owns each domain.
The foundation rung. Sources land on a known path without a project being opened for each one, failures and nonsense data both raise alerts, a named business steward can correct a record without a developer, and you know what the pipelines cost to run.
BI and FP&A analyst self-service
Analysts build on shared models, through a review path, under source control.
Shared domain models replace one-model-per-report. Metric definitions travel with the measure instead of living in a side document. Analysts change what they use daily without filing a ticket, and a peer reviews it the same day through Git.
Business user self-service
People outside the data team build and publish safely, and know how.
The capability exists and the business knows it exists. There is a defined place a business user is allowed to publish to, with the certification, permissions and review that make it safe for someone else to rely on.
AI-focused workstreams
AI tools answer over your models consistently and within permissions.
An agent pointed at your models returns the same number whoever asks and however they phrase it, and cannot leak past the asker's permissions — because the rules live in the model and apply through every path to it, including the agent's.
Which rung are you actually on?
Reading the ladder and placing yourself on it are different things, and teams reliably place themselves one stage high. The readiness assessment is nineteen questions that settle it — each with a concrete bar for what a Yes looks like, so the answer is not a matter of opinion.
Change log
This page is maintained. It changes when the practice changes, and every revision is listed here.
- 2026-10-29
- First version, lifted from the capability roadmap used in the services handout.