BI Pixie Workload in Microsoft Fabric
Generally Available

Proposed Fields

With the strategy set, select Propose fields. BI Pixie reads the semantic model, ranks its measures, columns, and filter values by the strategy, and shows you what the questions will be built from before a single question is written. You never have to hand-pick fields, but you always can.

On a semantic model with a saved benchmark, Propose fields waits until BI Pixie has loaded the saved strategy, and says so in place. Nothing else on the screen is disabled without a reason you can read.

The Proposal

The proposal arrives as three cards, directly under the strategy.

The fields your questions will be built from

The measures, columns, and filter values BI Pixie proposes, in a fixed order: your business domains first, then the remaining tables under Other tables. Inside each domain the fields cluster under the table they come from, and each table heading carries the field count, the usage evidence behind the table, and a control to fold the cluster away. Each field chip carries its own evidence, such as how often it is used or which reports show it. Remove anything you do not want tested, or select Add more fields on any section to open the ranked list of what else the semantic model offers. A field you add appears at the end of its table's cluster, where you can find it.

The fields a benchmark will be built from, grouped by business domain and table, with each measure and column shown as a chip. A line above the groups reads that click counts cover the last 90 days of activity in the reports BI Pixie tracks, and each field and each table carries its own count: 109 clicks on the Campaigns table, 20 each on Total Emails Sent, CTR, Total Open Rate and CTOR, 8 on Campaign, 11 on List Name, and 6 on Calendar.

A domain with nothing to show and nothing to add is not listed. The one exception is a domain you singled out in the strategy that contributed no fields, which stays on the card with a line saying so.

Questions built from your report visuals

The measures and breakdowns your tracked reports put on screen, asked as questions and graded against your data. Each row states the question, the slice it asks for, and the visual and page it came from; the report's name and the date BI Pixie last read it are stated once above the list where every row shares them, and on each row where they differ. The two cards are one benchmark: removing a field from the first card also removes the report questions that depend on it, which BI Pixie confirms first, naming what would be lost.

The questions built from report visuals, in two columns, each naming the visual and the report page it came from, with a note that every question comes from the report Marketing Campaigns. Some rows add that the question asks with the visual's own filter, and others that it asks across all data without it. A footer reads that the benchmark will ask 15 questions, 8 simple, 7 standard and 0 complex, above a Generate questions button.

BI Pixie reads a visual's field wells rather than reproducing its query. Where a chart's filter cannot be reproduced, the row says the question asks across all data without the visual's filter, so its answer will not match the number on the chart. Questions filter on text fields only. If no report on the semantic model is tracked, or no visual inventory has been captured yet, the card states which of the two applies.

The count, and Generate questions

The third card states how many questions the benchmark will hold and carries Generate questions, with a note that generating does not use a run. On a semantic model that already has a saved benchmark, BI Pixie asks before generating and names what will happen: it replaces the generated questions, including any you edited, and keeps the questions you wrote yourself. Canceling changes nothing. Once questions exist the proposal folds away; Show the fields and Hide the fields bring it back and put it away.

When Your Reports Change

When your reports change after a benchmark has been built, a notice reports it: "Your reports changed since this benchmark was built", with the visuals the benchmark tests that are no longer on your reports, and the visuals your reports show that the benchmark does not test. Select Propose fields again to rebuild the benchmark from what your reports show now. Past runs and their scores are kept.

What's Next