- Daily Success Snacks
- Posts
- 5 Brutal Truths About AI in Power BI
5 Brutal Truths About AI in Power BI
Copilot doesn’t fix Power BI. It accelerates whatever your model already is.

Read time: 2.5 minutes
Someone is expecting Copilot to perform magic by putting a stop to manual DAX. Moments later, they received a measure generated from a revenue definition that their organisation would NEVER use!
Debate begins to erupt...is Copilot in error or is the dataset invalid? The reality of the situation is simply that AI is not the cause but rather just the accelerant.
5 Methods for Resolving Problems Before Relying on AI in Power BI:
1) Copilot Will Not Resolve Unorganized Models, It Will Only Enhance Their Disorder
A disorganized semantic model will yield disorganization more quickly when an AI is involved.
The first solution will be to ensure the semantic model is clean:
Clear the house on your tables' names
Discontinue use of non-value-added columns
Ensure there is a description of measures
Reinforce the use of the star schema winning mode
Using a garbage semantic model will produce garbage AI results.
2) Inadequate Date Table Means Poor Time Intelligence
Provide a marked date table in order to use any time functions
The recommended solution will be:
Build a historical date table
Use a date column with no duplicates
Mark as a date table
Use the date table to build all relationships
Basic business intelligence knowledge should provide guidance here, but many don't follow it.
3) Bi-Directional Relationships may seem Smart, however, They Are Most Often Not
The use of bidirectional relationships will slow model performance, create ambiguity about the model, and mask design problems.
The solution is to use single-directional relationships wherever possible.
The use of the dimensional model will filter out loose fact records to produce the star schema-winning relationship design.
4) "If a Visual Loads, Then It Works" Will Produce Slow Dashboards
Without the Performance Analyzer Tool, you are left to guess.
The solution will be:
View→ Performance Analyzer
Record→Add Slow Visuals→ Fix
The faster dashboards render, the more trust people place in their data.
5) AI Produces Syntax and Not Strategy
AI has no knowledge of Revenue, Exclusions, or Fiscal Procedures.
The solution will be:
Create KPI definitions in writing
Use standardized logic(all Certified Datasets)
Review all AI generated measure
AI produces a draft of a measure that the user will approve as the final version.
💡Key Takeaway:
The presence of AI isn't going to result in the elimination of the role of a Power BI developer. The undisciplined Power BI developer who misuses AI technology will eliminate themselves.
👉 LIKE this if you know of people who have experienced an "incorrect" measure created by AI or "true" measure created by AI.
👉 SUBSCRIBE now to read about Power BI and analytics to help improve decision-making.
👉 Follow Glenda Carnate for best practices in governance and modeling that will enable you to survive through reorgs/AI implementations.
Instagram: @glendacarnate
LinkedIn: Glenda Carnate on LinkedIn
X (Twitter): @glendacarnate
👉 COMMENT with the most dangerous “working” measure you’ve ever inherited.
👉 SHARE this with a team treating Copilot like a modeling strategy.
Reply