AI Readiness

Copilot in Power BI, Fabric data agents, and MCP clients generate their answers from the metadata inside your semantic model: table and field names, descriptions, relationships, and the instructions the semantic model carries. Where that metadata is thin, an agent infers which field a question refers to and how to aggregate it, and returns a plausible wrong number rather than an error.

The AI Readiness page listing semantic models with their readiness score, benchmark score, last assessment date, and benchmark runs

AI Readiness measures how well your semantic models are prepared for Copilot in Power BI and for Fabric data agents, helps you close the gaps, and verifies the outcome. Open AI Readiness from the left sidebar of the BI Pixie Portal. The Overview page also carries an AI Readiness card with an Open AI Readiness button.

In This Guide

AI Readiness is two kinds of work, and this guide follows the same split the product does.

  • Assessments: score a semantic model from 0 to 100, read the findings, and track the score over time.
    • AI Provider: connect the AI that writes descriptions, AI Instructions, folder names, and benchmark questions for you.
    • Score Dimensions: the thirteen measurements behind the score, and why each matters to an agent.
    • Optimizations: the fixes BI Pixie proposes and writes into the semantic model once you approve them.
    • Schedules: assessments that run themselves, on the semantic models you choose.
  • Benchmarks: ask an AI real business questions about the semantic model and score its answers against your own data.
    • AI to Test: Power BI Copilot, a Fabric data agent, or a generic AI agent, and what each one needs.
    • Strategy: how many questions, which areas of the business, how demanding, and how often each one is asked.
    • Proposed Fields: the measures, columns, and report visuals the questions are built from.
    • Questions and Answers: review, edit, and add questions, and check every correct answer before the run.
    • Run and Results: what a run does, how to read the result, and how to repeat a benchmark fairly.

Before You Start

  • A BI Pixie account on any plan, including the free plan. Scoring is included on every plan, and the first score of any semantic model is free.
  • Fabric access to the semantic models you want to work on: your own workspace permissions at Contributor level or higher. BI Pixie reads the semantic model definition through the Fabric API with your identity, and can never reach a semantic model you cannot. AI Readiness is not admin-only: any member of your BI Pixie account can use it.
  • For AI-authored text, an AI provider. Everything that does not use AI works without one.
  • For a benchmark, whatever the AI you choose to test requires. Neither assessing nor optimizing needs the BI Pixie Dashboard.

The AI Readiness Page

The page header carries a Preview pill, because some optimizations rely on Microsoft's Prep data for AI, which is still in preview, and two buttons: Start assessment and New benchmark. If you are an account admin and no AI provider is connected, a one-line callout below the buttons carries a Set up action. Once a provider is connected, the callout no longer appears, and a small icon button beside the header buttons names the provider in its tooltip and opens its settings. Members who are not admins see neither the callout nor the button. On the Enterprise plan, one further line summarizes what runs automatically, such as "4 schedules running. Next run Monday 06:00", and links to Schedules.

Your semantic models

The rest of the page is Your semantic models, one row per semantic model you have assessed or benchmarked. The list is searchable by semantic model or workspace, sortable, and paginated at 25 rows. When the list spans more than one workspace, a Workspace dropdown narrows it to a single workspace. Three tabs above it show the same list through one kind of work at a time:

Tab What each row shows What a click does
Overview (the landing tab) The readiness score, the latest benchmark score, when the semantic model was last assessed, and how many benchmark runs it has. Lowest readiness score first. Opens the semantic model's own page. Two icon buttons on the row start a New assessment or a benchmark.
Assessments The readiness score, its change since the previous assessment, how many assessments exist, and the date of the latest. Nothing about benchmarks. Expands the row to list that semantic model's assessments. The row carries a New assessment button and an actions menu.
Benchmarks The latest benchmark score, which AI it was measured through, its change, and how many runs exist. Lowest benchmark score first. Semantic models with no benchmark runs are not listed. Expands the row to list that semantic model's benchmark runs. The row carries a benchmark button and an actions menu of its own.

On the Assessments tab, each row leads with a New assessment button and offers no benchmark action. Its actions menu holds two entries: View activities, which opens the semantic model's own page, and Delete assessments. On the Benchmarks tab, each row leads with one benchmark button whose label follows the semantic model's state: Run benchmark where a run exists to repeat, New benchmark where none does, Resume benchmark where you saved questions without running them, and View progress while a run is underway; the benchmark icon on the Overview tab's rows follows the same rule. The Benchmarks tab's actions menu holds View activities, Edit benchmark, and Delete benchmark runs. The sort control offers Lowest score first and Name A to Z, except on the Benchmarks tab, where the choices are Lowest benchmark first and Name A to Z.

To score many semantic models in one run, select them in the list and use the Assess button that names how many you selected. See Assessments.

A semantic model's own page

Every semantic model has a page of its own, reached from the Overview tab or from View activities in a row's actions menu. The header states its name, its workspace, its current score with the band in words, and when it was last assessed, and carries the New assessment button, the benchmark button, and the actions menu, including Delete all results. On the Enterprise plan the header also says whether assessments run for this semantic model on a schedule, with a control to add it to one.

The page for one semantic model, showing its readiness score, its assessment history with the optimizations applied, and its benchmark runs

Below the header, two collapsible sections, Assessments and Benchmarks, list every record newest first. A section with records opens; an empty one starts shut and, when opened, says what that kind of work does and offers to start it. A long history shows its first rows with a Show all control. Every record opens to its full result, and the breadcrumb above a result, AI Readiness / the semantic model / the date, links back to each level.

The BI Pixie Dashboard semantic model is never offered when you choose a semantic model to assess or benchmark. BI Pixie rewrites that semantic model on every dashboard update, so curation applied to it would not survive.

Free First Look and Tracked Items

The first assessment of a semantic model is free on every plan: each account may survey up to 500 distinct semantic models without any of them counting against the plan. A semantic model begins counting on its second assessment, its first optimization, or its first benchmark run, whichever comes first. At that point it becomes a tracked item, the same quantity your plan uses for reports with Pixies. A report and its semantic model are counted separately, even when the report is built on that semantic model. The portal states this at the claiming step, and if the action would exceed your plan's tracked items it is blocked with an upgrade prompt. The free first assessment is never blocked, and deleting results never returns a semantic model to the free state.

Plan Allowances

Plan Tracked items Assessments a day Benchmark runs Also
Free 1 report and 1 semantic model 10 1 in total Optimization of the one tracked semantic model, up to 10 times a day.
Standard 10 100 3 in total Up to 200 report viewers.
Pro 50 250 100 a month Up to 2,000 report viewers, the benchmark's Thorough and Custom question counts, in-report feedback, and embedded surveys.
Enterprise 500, or unlimited as part of your quote 1,000 Unlimited Unlimited report viewers and scheduled assessments.

On every plan you can build a benchmark end to end and inspect the answers BI Pixie computes; the number of runs is what differs by plan. See Plans and Billing for the full picture.

The portal runs the same AI Readiness as the BI Pixie Workload for Microsoft Fabric, with the same pages, the same scores, and the same benchmark. One detail differs, and it is noted where it applies: the portal's AI provider is usually your own Azure AI Foundry, OpenAI, or Anthropic account, where the Workload's default is the BI Pixie data agent in your workspace.

What's Next

  • Assessments, to score your first semantic model.
  • Benchmarks, to measure what an AI actually returns from your semantic model.
  • Add Pixies to Your Reports, so that scores and benchmark questions are grounded in real usage.
  • AI Assistants, to create the BI Pixie data agent and use it as your AI provider.