The Executive Insight
Most organizations still talk about “reports” and “dashboards” as if they are the product.
In Fabric, that mindset is already out of date.
The real product-- the thing that determines whether AI, Copilot, and self‑service analytics actually help or quietly damage trust, is the semantic model: the shared layer where you decide what a customer is, how revenue is calculated, which time logic is “official,” and who is allowed to see what.
Fabric didn’t invent semantic models, but it did something more dangerous: it made them central and easy to multiply. If you don’t have a strategy, you won’t end up with “more insight.” You’ll end up with three, five, or ten slightly different truths, now accelerated by AI and shared faster than ever.
Issue #6 is about that layer: not how to click through and build one more model, but how to think about semantic models as the core of your data product strategy in Fabric.
The Quiet Problem: Too Many “Versions of the Truth” Wearing Different Clothes
In mid‑market companies, I keep bumping into the same pattern when Fabric enters the picture:
- A lakehouse or warehouse is created.
- Fabric generates a default semantic model.
- A team builds a “quick” report on top of it.
- Another team copies the model, tweaks it a bit, and publishes their own.
- A third team builds a brand‑new model on the same tables because “it’s faster than understanding the old one.”
Within a year, you have:
- 4–5 models all claiming to represent “Sales,” each with different grain, measures, and security.
- Power BI reports wired to each of them.
- Copilot and agents now sitting on top of this mess, trying to answer questions across inconsistent semantics.
The metrics look similar. The numbers are close. The definitions are not.
From a distance, it looks like progress. Up close, you start to see the cracks: executive meetings where Sales and Finance bring two dashboards to the table and both are “right according to their model.”
Fabric didn’t cause that. It just made it much easier to get there.
Fabric’s Design Choice: More Power, More Choice, More Responsibility
Microsoft’s own guidance is clear: in Fabric, the semantic model is not an afterthought. It is the central contract between data and the rest of the business.msfabric+2
Several signals point in the same direction:
- Guidance now recommends unified, shareable semantic models over one‑off report models.
- Default semantic models are being decoupled from their source items and treated as first‑class, independent assets.
- Semantics are being wired directly into Fabric IQ and data agents to give AI a stable ontology to reason over.
This is good news if you’re willing to treat semantic modeling as design, not plumbing.
It means:
- One model can serve dozens of reports, teams, and tools.msfabric+1
- AI can anchor itself in well‑defined entities and KPIs, not raw tables.refactored+1
- Governance can operate at the model level instead of chasing down every individual report.
But it also means this: every undisciplined model you allow into production is a new fork of reality.
A Pattern From the Field: When “Default” Becomes “De Facto”
One concrete pattern I see a lot in Fabric pilots:
- A lakehouse or warehouse is created.
- Someone uses the default semantic model because “it’s already there.”
- They drag some fields, create a couple of measures, and build what becomes the most‑used report in the company.
Months later, problems appear:
- The model started as a convenience layer, not a carefully designed star schema. It has awkward relationships, many‑to‑many joins, and performance issues.
- Security was “good enough for now,” then quietly became the de facto security model for sensitive data.
- Other analysts started building on top of it because it was the easiest thing to connect to.
Now you’re stuck:
- You can’t change the model without breaking downstream reports.
- You can’t easily replace it because 30+ artifacts are wired into it.
- You can’t certify it with a straight face because you know it wasn’t designed to be the enterprise standard.
What started as the default silently became the standard.
The hard truth: you don’t get to avoid semantic design. You only get to choose whether you do it intentionally up front, or painfully later under load.
The Strategic Shift: Think “Platform Model,” Not “Report Model”
If you want Fabric to be more than a slightly better reporting engine, you have to start thinking in terms of platform models:
- Models that are designed to be shared via Build permissions, not owned by one report developer.
- Models that are certified and discoverable as the “official version” for a domain.msfabric+1
- Models that are understood by humans, not just engines--clear naming, clear grain, clear role‑level security.
Some of the most forward‑thinking teams I work with have adopted a simple philosophy:
“If a semantic model isn’t good enough to be shared, it isn’t good enough to exist in production.”
That doesn’t mean every experiment has to go through a committee. It means there’s a clear line:
- Sandboxes where analysts can explore, prototype, and throw things away.
- Platform models that are curated, governed, and intentionally reused.
Fabric’s feature set (DirectLake, shared models, Fabric IQ) strongly encourages that separation.refactored+2 Ignoring it doesn’t just cost performance; it erodes trust.
What This Means for AI, Again
In Issue #5, we talked about Copilot not being your analyst, but your mirror. The semantic model is the mirror frame.
Copilot, data agents, and Fabric IQ don’t “understand” your business in the abstract. They understand:
- The entities, relationships, and measures you’ve exposed through semantic models.
- The permissions and domains laid out by your governance setup.enterprise-knowledge+2
If those models are:
- duplicated,
- inconsistent,
- or lacking in basic clarity,
then AI will behave the same way--duplicated, inconsistent, and unclear.
The promise of Fabric IQ, a semantic intelligence layer over your business--only holds if you feed it a coherent ontology.
That work is not glamorous. It’s also exactly where long‑term advantage comes from.
Hope for Lean Teams: You Don’t Need a Perfect Enterprise Model
This can sound intimidating for lean SMB teams who are already stretched thin. The good news is you don’t need a perfect global enterprise model to start winning.
You need one well‑designed domain model that proves the point.
The patterns I see working well in $50M–$100M organizations look like this:
- Choose a critical domain, often Revenue, Operations, or Inventory-- where misalignment is currently expensive.
- Invest in a single, high‑quality semantic model for that domain: star schema, clear measures, row‑level security you’re proud of.
- Make that model the only certified source for that topic. Everything else is clearly labeled as exploratory or legacy.
- Wire Copilot, agents, and self‑service reporting only to that model at first.
What happens?
- Executives start to notice that “the numbers finally match.”
- Analysts spend less time arguing about data shape and more time asking better questions.
- AI outputs stop feeling like a parlor trick and start feeling like an accelerant on a known truth.
At that point, expanding to a second and third domain stops being a theoretical debate. You have proof.
Where I Fit In (For Partners and Leaders)
Most Fabric conversations I see are still dominated by:
- “Which workload should we put in a lakehouse vs warehouse?”
- “Should we use Mirroring, Shortcuts, or Pipelines for this source?”
- “How do we turn on Copilot for our tenant?”
Those are important, but they’re not enough.
Without a semantic strategy, every other decision is just rearranging the same confusion in new tools.
I work with:
- Partners who need a semantic and governance backbone for their Fabric projects, so they’re not shipping beautiful reports on top of unstable definitions.
- CIOs, CDOs, and CTOs who want “single version of the truth” to be more than a slide in a board deck.
- Business leaders in F&B, healthcare, and energy who are tired of executives walking into meetings with two dashboards and three definitions of margin.
My focus is simple to state and hard to execute:
Design semantic models that your business, your AI, and your governance can all live with—at the same time.
If you’re building on Fabric and you know that reports, AI, and governance are all pulling on the same thread, the semantic layer, but you don’t have the time or appetite to untangle it alone, that’s where partnership makes sense.
Isaac Truong | Founder, Allston Yale Strategic Data Architecture & AI Readiness for the Enterprise Mid‑Market
If you’re a partner or leader who wants Fabric to be a platform, not just a reporting tool, and you’re ready to treat semantic models as your real product, let’s talk.
What domain would you want to fix first if you could snap your fingers, Revenue, Operations, Inventory, or something else?
Isaac Truong | Founder, Allston Yale
Enterprise-grade analytics for $50M–$100M SMBs
Power BI | Fabric | Azure | Data Strategy
📅 Book a 20-min Fabric diagnostic →
📧 Subscribe to get Friday Fabric Facts in your inbox (plus early access to templates) 💼
LinkedIn: Connect with me for daily Fabric tips
Friday Fabric Facts #6: Originally Posted on LinkedIn, March 6, 2026
