Originally published August 30, 2026. Updated October 2026 to reflect current Fabric IQ terminology and capabilities.
Is Your Power BI Semantic Model Ready for AI?
Key takeaway: A Power BI semantic model can be perfectly ready for dashboards and still not be ready for AI. AI readiness means reducing ambiguity by making business terminology, authoritative metrics, descriptions, relationships, analytical rules, and other context explicit enough for Copilot and other AI experiences to interpret the business reliably.
A Big Change for Power BI Teams
A Power BI semantic model can be excellent for dashboards, yet a poor foundation for Generative BI. That might sound strange at first. If the measures are correct, the relationships work, the reports render, and business users trust the dashboards, what else is there to do?
Quite a bit, actually. The reason is simple:
Humans bring business knowledge to a semantic model. AI tools like Copilot and Fabric data agent lack this business knowledge.
A report developer may know which of several similar measures is the authoritative one. They may also know:
OrderDate should be used for one analysis while ShipDate should be used for another.
That a column named Status refers to customer status rather than order status.
Which tables are legacy, which measures should be ignored, and which business rules exist only in the heads of subject-matter experts.
A dashboard can work perfectly well because the person who built it already knows all of that. An AI system has to infer it. That creates a new question for Power BI teams:
“Is our semantic model clear enough for AI to use reliably?”
Dashboard-Ready Is Not the Same as AI-Ready
For years, Power BI teams have optimized semantic models primarily for two consumers:
Report and dashboard developers (yes, be honest)
Business users interacting with reports and dashboards
However, Generative BI introduces another consumer:
AI tools like Copilot and Fabric data agent
That changes the design problem. A Power BI developer can work around ambiguity. In fact, the more experience they have, the more they can work around it.
AI agents are not like that at all. For example, think of a Fabric data agent as a bright, confident, and motivated university hire. Yes, they will get a ton of work done fast.
But without the required business knowledge, most of that work will be well-meaning and wrong.
Suppose your Power BI semantic model contains these measures:
Revenue
Net Revenue
Revenue Adjusted
Revenue Final
A Power BI developer will know exactly which one belongs on an executive dashboard. But what happens when a business user asks:
"What was revenue last quarter?"
Which measure should the AI agent (i.e., the bright, confident, motivated university hire that doesn't know the business) choose?
Now consider a Power BI semantic model containing:
Customer Name
Account Name
Client Name
Are those three different concepts? Three names for the same concept? Does one apply only to a particular business process?
Again, the Power BI developer will know the answer.
Without sufficient business knowledge (i.e., context), the AI agent will have to guess. Or worse yet, the AI agent will make something up (i.e., hallucinate).
And nothing will kill your Generative BI initiative faster than AI agents that hallucinate answers for executives.
This is why AI readiness for Power BI teams is really an ambiguity problem.
The more ambiguity the semantic model contains, the more room AI has to make the wrong choice - or make stuff up out of thin air.
What Should "AI-Ready" Mean?
I don't think "AI-ready" should mean that a model passes a single checklist or uses every available Power BI and Fabric IQ feature. A better definition is:
An AI-ready Power BI semantic model gives the Generative BI system enough clear, trustworthy business context to choose the right data and interpret it correctly.
That means the model should help AI answer questions such as:
Which column represents the business concept the user mentioned?
What business terminology should be interpreted in a particular way?
Which measure represents the official version of this metric?
Which objects are irrelevant or potentially confusing?
How are two concepts related?
Which date should be used?
This goes beyond whether the semantic model is technically valid. This goes beyond making sure your DAX code is efficient.
It's about whether the model communicates meaning.
Start with Naming
One of the easiest places to begin is naming. Consider a column called:
Amt
A Power BI developer will immediately understand what it means. Hopefully, the AI agent's training data will allow it to interpret it correctly - but maybe not. Now compare that with:
Sales Amount
The second name provides more evidence. The same principle applies to measures.
Your finance team might be totally comfortable using GM% in PowerPoints, reports, and dashboards.
Again, you have to hope an AI agent has been trained on enough data to interpret this correctly. A far better idea is to make everything crystal clear:
Gross Margin %
That does not mean every name needs to be painfully long and explicit (although it's not a bad idea, to be honest). Mostly, it means names should use consistent business terminology.
If the business says Customer, avoid using Client, Account, and Buyer interchangeably - unless those terms truly represent different concepts.
Consistency reduces ambiguity. And reducing ambiguity is one of the central goals of preparing a semantic model for AI.
Make the Authoritative Metrics Obvious
DAX measures deserve special attention because they often contain the most important business logic in a Power BI semantic model.
The problem is that many mature models accumulate measures over time. You might find:
Multiple versions of the same KPI
Measures with nearly identical names
Old measures that are no longer used
Temporary measures created for specific reports
Measures whose intended use is obvious only to the original developer.
This may not hurt a dashboard because the correct measure is already on the page. But a Generative BI system answering an open-ended question has to choose.
That makes governance around measures critical for your success with Generative BI.
If the business has an official definition of revenue, margin, churn, active customer, or any other important metric, the AI agent should not have to guess which measure represents it.
An AI-ready semantic model should make authoritative metrics easy to identify and prevent confusion with competing alternatives.
Descriptions Are No Longer Optional Documentation
Power BI semantic model descriptions are another place where AI changes the value equation. Historically, descriptions were easy to skip. Power BI semantic models work just fine without them.
The report still rendered.
The DAX still calculated correctly.
So, documentation often became something teams planned to add later.
Generative BI makes thorough descriptions an absolute requirement. Thorough descriptions can explain:
What a measure means
When it should be used
When it should not be used
How it differs from a similar metric
What a column represents in business terms
What assumptions are embedded in the calculation
Compare a semantic model with a Revenue measure with no description to a semantic model with a Revenue measure with the following description:
Net product revenue after discounts and returns. Use this measure for standard executive sales reporting. Excludes freight and tax.
That description provides context that the measure name and DAX expression alone may not provide. Your descriptions must reduce uncertainty for the AI agent. A description such as:
Revenue: The amount of revenue.
Does not accomplish much.
Don't make this mistake. Enriching your Power BI semantic models sufficiently for real-world Generative BI will be a substantial effort for most Power BI teams.
Leveraging multiple semantic models is essential for delivering valuable Generative BI to your organization. Every semantic model used for Generative BI needs common names and thorough descriptions.
For example, this is absolutely required when you layer a Fabric IQ ontology on top of your Power BI semantic models (and you will want to do this).
Watch for Hidden Complexity
When your Power BI team gets serious about Generative BI, here's the first question you should ask:
"If we hired a new Power BI developer, how quickly and thoroughly could they get up to speed by only using the semantic models as-is?"
Don't feel bad if the answer is, "It would take a long time, and it wouldn't be a very thorough understanding."
Power BI teams often have many opportunities to improve their semantic models. Many Generative BI problems hide in these missed opportunities. For example:
Inactive objects may remain visible
Similar fields may exist in several tables
Technical columns may sit beside business-facing fields
Important business rules may exist only inside complex DAX
Model-specific terminology may differ from the language users actually use
The model may expose far more objects than a business user would ever need
None of these automatically make the semantic model bad. But collectively they increase the amount of interpretation required.
For a human developer, that interpretation may be routine.
For an AI agent, every unnecessary choice is another opportunity to select the wrong thing.
Sometimes, preparing a model for AI means adding context.
Sometimes, it means removing noise.
Relationships Matter Too
Relationships are often discussed as a technical modeling concern (e.g., foreign key relationships). However, relationships often define the business's most important concepts. A relationship may tell you that:
A Customer places Orders
A Product belongs to a Category
A Distributor operates in a Geography
A Transaction occurred on a particular Date.
The analytical engine needs the technical relationship. Real-world Generative BI must have the conceptual meaning behind the business relationships.
This becomes especially important when the model contains multiple possible paths or when a relationship is technically correct but not obvious from the names of the objects involved.
The lesson is the same:
AI works better when the model makes the business easier to understand.
Some Questions Deserve a Trusted Answer
Power BI's new "Prep data for AI" feature also supports specifying Verified Answers. This addresses another important problem in Generative BI.
Not every business question should be treated as an entirely open-ended generation task.
Some questions are common. Some are important. Some have answers the business has already agreed upon.
For those questions, consistency matters far more than creativity.
Verified Answers allow semantic model authors to associate common questions with curated responses. That creates an interesting design choice for BI teams:
Which questions should AI reason through dynamically, and which questions should have an approved answer?
That's not just a Power BI feature decision. It's a governance decision.
And I expect questions like this to become increasingly important as Generative BI moves into real-world production use.
AI Instructions Fill-In What the Model Can't Say
Some business knowledge does not fit neatly into a table, relationship, measure, or description. Here are some examples:
"When users say customers, they mean active paying customers unless they specify otherwise."
"Do not compare this KPI across regions because the calculation methodology differs."
"Use fiscal year rather than calendar year for executive performance questions."
"Prioritize net revenue over gross sales when both are available."
This is where Power BI's "Prep data for AI" lets you add AI instructions directly to your semantic models.
AI instructions allow the semantic model developer to provide business logic, terminology, and analytical guidance that AI agents such as Copilot can use to interpret questions.
Again, the broader principle matters more than the feature itself:
Generative BI needs business context that traditional analytical metadata may not fully capture.
"Prep data for AI" gives Power BI teams additional places to provide that context. But those capabilities are most useful when they build on a well-designed semantic model.
They don’t replace one.
A Quick AI-Readiness Assessment
If I were evaluating an existing Power BI semantic model for Generative BI, I would start with a few questions.
Terminology - Would a business user understand the names of the important tables, columns, and measures?
Consistency - Does the model use the same terms consistently for the same business concepts?
Metrics - Are the authoritative measures obvious, or are there competing versions that could confuse AI?
Descriptions - Do the important objects explain their business meaning and intended use?
Complexity - Are unnecessary or technical objects creating noise?
Relationships - Does the model clearly represent how the major business concepts connect?
Business knowledge - Are important definitions and rules encoded somewhere AI can actually access them?
AI configuration - Could an AI data schema, verified answers, or AI instructions reduce remaining ambiguity?
If several of those questions are difficult to answer, the model may be dashboard-ready without being Generative BI-ready.
And that’s exactly the gap Power BI teams should start looking for.
The Bigger Opportunity for Power BI Teams
This is why I believe Power BI professionals are well positioned for Generative BI. The work is not completely foreign. Power BI teams already understand:
Dimensional relationships
Business terminology
Business metrics
Analytical logic
Data quality
Generative BI simply makes it more important to make that knowledge explicit and machine-readable.
The semantic model is no longer only a layer between data and dashboards. It's becoming a layer between AI agents and your business.
That changes what "good modeling" means.
Greetings! I’m Dave Langer
Author | Microsoft MVP | LinkedIn Top Voice
I help Power BI teams move past dashboards to Generative BI using Microsoft Fabric and Fabric IQ.
If your organization is serious about enterprise-class conversational analytics, we should talk.