BI Pixie Workload in Microsoft Fabric
Generally Available

Questions and Answers

Select Generate questions and BI Pixie writes the benchmark: one question per slot in the proposal, each with the DAX query that computes its correct answer from your data. With an AI provider connected, the questions are phrased naturally; without one, they follow fixed templates. The provider receives field names and the shape of each question, never the answer, so the answer key cannot leak into the test. The questions land in the Review the questions card, open, with Run benchmark at its foot.

Review Each Question

Each row states the question, its complexity level, and where it came from: a business domain, or the report visual it reproduces. Expand a row to see how it will be graded:

The Review the questions list, with eight proposed questions. Each row carries a checkbox, its complexity level, the question text, and an arrow that expands it. Below the list are Add your own question, a note that eight questions asked three different ways plus up to eight answer checks need no Fabric capacity, and the Run benchmark and Save for later buttons
  • Check answer computes the expected value from your data and shows it as the Ground truth, so you can verify every answer before the run.
  • Edit the query. The grading DAX is yours to change. An edited row is badged Edited and must be checked again before running. Use BI Pixie's query again restores the original.
  • Edit the wording. Rephrasing changes only the text sent to the AI. The answer check still uses the query.
  • Question variations. When the strategy asks each question a different way, every variation is listed under its question and can be edited before the run.
  • Remove a question. A question you remove is gone from this benchmark only; Propose fields again can always rebuild the list.

Add your own question

Select Add your own question, write the question and its ground-truth DAX, check the answer, then select Keep question. Your rows are badged Yours. When you generate new questions, BI Pixie replaces only the generated ones and keeps the questions you wrote yourself, and the confirmation says so.

Saved Automatically

Every change to the questions is saved as you make it, without a blocking spinner. Save for later confirms it in words: "Saved. You can pick this up from AI Readiness whenever you are ready." The semantic model's row in Your semantic models then shows a draft line, such as "Draft: 14 questions saved 3 days ago", and a Resume button that opens the questions exactly as you left them. If a save fails, BI Pixie says so and offers to try again before you leave.

There is one saved benchmark per semantic model. Generating new questions replaces it, and BI Pixie names what will be replaced first; opening New benchmark and leaving replaces nothing.

Before the Run

Before the run starts, BI Pixie verifies which questions can be scored. A row that cannot be scored is badged Can't be scored with the reason, such as a busy capacity, a grouping with no effect, or no data in the chosen period. A confirmation lists them, and Run anyway starts the run. An unscorable question still counts in the score's denominator, so the result never looks better for having skipped it.

If BI Pixie has improved how it writes questions since yours were saved, a line above the list says so, and Propose fields again rebuilds them the new way. Questions edited after a run are marked in the history, so an edited benchmark is not mistaken for a fair comparison.

If the Saved Questions Cannot Be Opened

If BI Pixie cannot read your saved benchmark when you open it, the screen says so, states that nothing has been changed or removed, and offers Try again. It never silently hands you an empty builder. In the rare case that the saved questions are gone, the screen says that too, notes that your completed runs and their scores are not affected, and offers Set up a new benchmark.

What's Next