Feature

Introducing AI Readiness in BI Pixie

BI Pixie now assesses whether your Power BI semantic models are ready for Copilot and Fabric data agents, drafts the context they are missing, and creates benchmarks to test your AI against answers computed from your own data. The first assessment of up to 500 semantic models is free on every plan.

The AI Readiness page running inside Microsoft Fabric as the BI Pixie Workload, with the Fabric navigation rail and ribbon around it, listing four semantic models with their readiness score, benchmark score, and last assessment date
BI Pixie AI Readiness Power BI Power BI Copilot Fabric Data Agents AI Assessments AI Benchmarks Prep Data for AI Semantic Model

For two years BI Pixie has measured how people use your Power BI: who adopts a report and who stops coming, what they click, how satisfied they are, how fast reports load, and who saw what. Today we release the biggest expansion to BI Pixie to help you scale your AI investment and take your Power BI semantic models to production with Copilot and Fabric data agents. BI Pixie assesses whether your semantic models are ready for AI, drafts the context they are missing, and creates benchmarks to test your AI against answers computed from your own data.

Along with the expansion, we now offer a Free plan that assesses up to 500 semantic models, and a simplified flat pricing with no usage meters.

AI Readiness is available in BI Pixie Workload for Microsoft Fabric or in the cloud version.

Your semantic models already hold what AI needs

You have spent years building the dashboards and operational reports your organization runs on.

Then AI arrived, and with it a great deal of noise and confusion. You have read that dashboards are dead, that nobody will open a report again, and that everything will happen in a chat box. Real concerns were creeping in. What is the future of BI professionals in a world with no dashboards?

As time passes and AI evolves with it, you discover that your expertise is needed more than ever. AI needs business context, and your semantic models are the best foundation. The years you invested in building semantic models will make AI work on your data. You are now more essential than ever.

But how do you take AI to production? More and more requests land on your desk: connect our data to Copilot, stand up a data agent, let people ask questions in their own words. Make it work. Make it accurate. And it usually arrives with a date attached.

Three problems between you and production

A semantic model that has served people well for years can still require extensive effort before an AI agent can be trusted to answer from it.

A wrong answer by AI looks like a right one. An agent that lacks context does not report an error. It returns a confident answer that may be wrong, and the first warning is a number nobody recognizes in an executive review. And even worse, trusting the answers without catching the error can be devastating to your business.

How do you get AI Readiness at scale? Scoping the context down to what matters most for AI and resolving ambiguity takes many iterations on a single semantic model. Your organization may have hundreds of semantic models that are now being considered for AI, and the teams that can properly curate and optimize the semantic models are scarce.

Stay correct as semantic models change. A new measure without a description in the semantic model can break an agent that answered correctly last month, and nothing flags it before an executive reports it. How do you measure the accuracy of the agents and detect regressions and drifts?

What we announce today with the release of AI Readiness in BI Pixie is the solution to these challenges, and we think it can take your AI to production.

Assess a semantic model

You point BI Pixie at a semantic model and it comes back with one AI Readiness score from 0 to 100 and a list of what is holding that score down. Thirteen dimensions sit behind it, running from description coverage and colliding names to the AI data schema, AI Instructions, and implicit measures. BI Pixie reads the definition through the Fabric API using your own identity, so it can never see a semantic model you cannot.

Score Dimensions explains what each dimension measures and why it matters to Copilot or an agent, and Assessments covers running one and reading the result.

An AI Readiness assessment result in the BI Pixie Workload for Microsoft Fabric. A dial shows a readiness score of 75 with the verdict Needs curation, beside thirteen dimension tiles that each carry a count and a percentage: AI data schema, AI Instructions, Description coverage at 6 percent, Name disambiguation, Field clarity, Indexing limits, Field defaults, Display folders, Implicit measures, Model structure, Time window safety, PII and RLS safety, and Verified answers. A footer reads eleven tables and 216 fields analyzed.
You point BI Pixie at a semantic model and get one AI Readiness score with the thirteen dimensions behind it, so you can see exactly where improvement is needed.
The page for one semantic model in the BI Pixie Workload for Microsoft Fabric, headed Marketing Campaigns, reading 75 Needs curation and last assessed 8/24/2026. An Assessments section lists four records, each with its score, the change against the record before it, the number of findings, what was found, and what was applied, including field descriptions and 47 display folders. A Benchmarks section below lists eight runs, each with its percentage, its change, the AI that answered, and a breakdown of correct, wrong, unstable, and not scored questions.
You can run multiple assessments for the same semantic model, and watch if your AI Readiness improves. This semantic model went from 70 to 75 across four assessments.

Optimize your semantic models for AI

Most of what an assessment finds is missing or ambiguous context: descriptions nobody ever wrote, an AI data schema that still exposes helper tables, the instruction an agent would need to pick the canonical revenue measure. Writing all of that by hand takes time. Let’s take the AI data schema as an example. You can use Power BI Desktop or the Power BI service to define the AI data schema using the “Prep data for AI” feature. But how would you choose which fields to include when you may be dealing with thousands of fields? This is where BI Pixie comes to the rescue. If your reports have Pixies, BI Pixie tells you which fields get the highest number of clicks and which are used the most in visuals. With this information, BI Pixie recommends the fields to select.

For some of the recommendations, BI Pixie allows you to select an AI provider that will help you generate the AI Instructions and descriptions for fields. All the proposed optimizations can be edited, and nothing changes until you approve them.

Optimizations covers the fixes BI Pixie proposes, and AI Provider covers connecting one.

The Optimize semantic model panel in the BI Pixie Workload for Microsoft Fabric, introduced by the line BI Pixie can make these changes to your semantic model for you. Cards offer Curate the AI data schema with a Review fields to exclude button, AI Instructions for Copilot and data agents with a Suggest AI Instructions button, Table descriptions noting eleven of eleven tables have no description, and Field descriptions covering 163 fields, grouped by severity with a checkbox on each table.
BI Pixie can recommend which fields to select for the AI data schema based on usage and lineage data. You can ask it to write AI Instructions and descriptions for tables and fields. You can review and edit it before it is saved.
The AI data schema review in the BI Pixie Workload for Microsoft Fabric, headed AI data schema with a Pixies inside badge and reading 180 fields in the AI data schema, 21 excluded, 0 changes pending. Tabs count All fields at 201, In schema at 180, Excluded at 21 and Recommended changes at 147, beside Set recommendations and Revert buttons. Each field carries its own recommendation and the usage behind it: Calendar / Date reads that users select this field 6 times and we recommend keeping it, marked On reports with 6 clicks on 9 report visuals in the last 90 days from your tracked reports, while Calendar / Week Day is marked Not on any report with 0 clicks and recommended for exclusion.
You see the clicks behind every recommendation. Calendar / Date is kept because your users picked it 6 times across 9 report visuals in the last 90 days, and Calendar / Week Day is proposed for exclusion because nobody touched it at all. In the BI Pixie Workload, inside Microsoft Fabric.

Benchmark the AI on answers you already know

An assessment tells you about the semantic model. A benchmark tells you about the answers, and the answers are what your stakeholders will judge.

A benchmark is an automated test you run to measure how accurately your AI answers from your semantic model. It comprises a set of business questions, each paired with the correct answer computed from your own data.

When the benchmark is ready, BI Pixie puts the questions to the AI you chose to test, grounded in your semantic model. It then checks every answer that comes back against the answer it prepared. What comes out is a score for how correctly that AI answers from your semantic model.

Building an effective test on your own is not easy, and it costs many hours, if not days. To help you scale that effort and automate the test, BI Pixie walks you through a series of steps.

The step that asks which AI you want to test, in a new benchmark in the BI Pixie Workload for Microsoft Fabric. Above it, a What gets tested card names the workspace and the semantic model. Three options follow, each with its own explanation and an Under the hood link: a Fabric data agent you already have on this semantic model, Power BI Copilot, and a generic AI agent that BI Pixie emulates by reading your semantic model's metadata.
You choose which AI to run the benchmark on: a new or existing Fabric data agent; Power BI Copilot; or a generic AI agent.

The first step is the strategy: how many questions the benchmark holds, which AI it tests, and how demanding those questions are. BI Pixie then proposes the fields that carry the most meaning in the semantic model, and draws question ideas from the visuals your reports already build on those fields. Every recommendation is yours to tune before the questions are generated.

The Benchmark strategy panel in the BI Pixie Workload for Microsoft Fabric, summarized as 10 questions, last month, all domains, usage-first tables, includes undocumented fields, balanced mix, asked 3 different ways. How many questions offers Quick check at about 10, Standard at about 15, Thorough at about 30, and Custom. How many times to ask each question is set to 3, with a choice between asking each question a different way and asking it the same way. A third card chooses whether later runs are scored against the original answers or against answers recalculated from your data.
Before BI Pixie generates the questions for the benchmark, you can set its strategy and define how demanding the test should be. For example, you can define the complexity of the questions and the number of variations of the same question you want BI Pixie to generate.
The fields a benchmark's questions will be built from, in the BI Pixie Workload for Microsoft Fabric. Notices say the benchmark leaves out 21 fields that the AI data schema hides from agents, includes 10 questions built from fields on your report visuals, and covers June 2024. Below, the fields are grouped by business domain and table: Campaign Performance and Engagement holds Campaigns with five fields and Calendar with two, and Audience and List Management holds Members and Lists with four fields and Lists with one. Each measure and column is a chip, and the described ones carry a Described mark.
BI Pixie proposes the fields in your semantic model that would be good candidates for the questions. It ranks the fields by real usage once your reports have Pixies, so the benchmark asks about the numbers your organization runs on.

Once BI Pixie generates the questions in the benchmark, you can edit any of them, along with the DAX query that defines the correct answer. After you approve the questions and DAX queries, you save it and can run it at any time.

The next time you run the benchmark, after a round of optimizations or major changes in the semantic model, BI Pixie compares the result with the run before, so a regression caused by a change to the semantic model can be detected quickly rather than as a support ticket three months later.

The Review the questions list in the BI Pixie Workload for Microsoft Fabric: eight questions, updated Aug 24, built from the tables and fields your users actually click, covering June 2024. Each row carries a checkbox, a complexity mark of Simple or Standard, 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.
BI Pixie generates the benchmark questions for your approval. You can review and edit the questions, the ground truth answer and the DAX query behind it.

A great deal of thought has gone into keeping that comparison honest. How do you ask the same questions of the same answers when the data underneath keeps moving? BI Pixie handles it in two ways. The first is that every correct answer can be recomputed from current data before a run, using the DAX query saved with each question. The second is that the questions are scoped to a time period that has already ended, where the data is unlikely to change further.

AI to Test covers what each choice requires, Questions and Answers covers reviewing and editing the questions, and Run and Results covers reading a result and repeating a benchmark fairly.

A benchmark result in the BI Pixie Workload for Microsoft Fabric, headed Checked 8/20/2026 via a generic AI agent. A 58 percent correct score sits above a strip of per-question marks and the line seven of twelve questions answered correctly. An About this run block lists who answered, the questions asked and scored, the question mix, the reporting period, what the AI could see, and which correct answers were used. Below it, a control chooses between the original answers and answers recalculated from your data, and a Weak spots section begins.
After you run the benchmark, BI Pixie shows you the accuracy level of your AI.
The weak spots of a benchmark run in the BI Pixie Workload for Microsoft Fabric. Three questions are listed as Wrong or Unstable, each naming the report visual and the page it came from, including one that asked across all data without the visual's own filter. An All questions section below shows every question with its verdict, its complexity, and its source visual, under a note that the questions were asked three different ways.
You can review each of the answers that AI returned. A wrong answer is an opportunity for improvement in the semantic model or in the benchmark.

A Free plan, and pricing that does not meter you

Two more things ship alongside AI Readiness.

Starting from today, there is no longer a 14-day trial period. You now have a Free plan!

Every BI Pixie account starts with the Free plan, in BI Pixie Cloud and in the Fabric Workload alike, and no credit card is required. You get an AI Readiness assessment of up to 500 distinct semantic models, one tracked report to measure usage, one tracked semantic model to optimize for AI as many times as you need, and one benchmark run on that semantic model.

Pricing is flat, by tracked items. A tracked item is a report you add Pixies to, or a semantic model you are preparing for AI. Your plan includes a set number of tracked items and nothing else is billed: no per-user fees, no usage billing, no overage charges. Pricing no longer moves with how much your reports are used.

PlanPriceTracked itemsBenchmark runs
FreeFree1 report and 1 semantic model1 in total
Standard$149/mo103 in total
Pro$499/mo50100 per month
EnterpriseFrom $30,000/yr500, unlimited availableUnlimited

Standard and Pro are two months cheaper if you pay annually. The Enterprise plan adds AI Readiness assessments on a schedule you set, so drift in a semantic model gets caught without anyone remembering to look. The pricing page carries the full comparison and the fair-use allowances.

What’s next

Learn more about AI Readiness in BI Pixie.

Start BI Pixie FREE plan today or if you prefer to run it on Fabric, ask your Power BI admin to approve the BI Pixie workload in the Fabric Workload Hub.