Use AI Run by BI Pixie for AI Readiness
BI Pixie can run the AI for AI Readiness for you. With AI run by BI Pixie, BI Pixie writes descriptions, AI instructions, display folder names and benchmark questions, and checks benchmark answers, without an AI service of your own. You do not need your own account or API key with an AI company. In AI setup, this choice is labeled BI Pixie AI.
BI Pixie runs the AI for AI Readiness on Microsoft Azure OpenAI in the same geography as your account's data: in the United States for accounts in the United States, and in the EU Data Boundary for accounts in the EU. BI Pixie uses one AI model, gpt-6-luna, and changes it only on purpose, after testing the replacement, and never automatically.
You decide whether to use AI run by BI Pixie. You can use your own AI instead, as described in Set Up Your AI for AI Readiness.
What BI Pixie does with the AI it runs
When BI Pixie AI is your AI in AI setup, BI Pixie does this AI Readiness work with it:
- BI Pixie writes descriptions, AI instructions and display folder names for the semantic models you optimize.
- BI Pixie writes benchmark questions in natural wording, and uses the AI as a second check on whether each benchmark answer matches the correct answer BI Pixie calculated from your own data.
- BI Pixie confirms whether two similarly named fields are really duplicates during an assessment.
- BI Pixie runs the AI round of a data agent assessment.
AI run by BI Pixie is only ever your AI. BI Pixie checks benchmark answers with it, so it cannot be tested as a generic AI agent in a benchmark or an A/B test. In the list of AIs a benchmark may test, the BI Pixie AI row appears closed, with the reason beside it.
Start using AI run by BI Pixie
If your account has saved no AI of its own, BI Pixie AI is already your AI and there is nothing to set up. BI Pixie can write text and check benchmark answers from your first visit.
An account that saved its own AI before BI Pixie AI was added to AI setup keeps that AI. BI Pixie AI appears as one more choice when an administrator selects Change on the Your AI row in AI setup. To use it, an administrator selects BI Pixie AI and then Use BI Pixie AI.
BI Pixie runs AI for AI Readiness only for accounts whose data is held in the United States or the EU. For an account whose data is held elsewhere, AI setup reads "BI Pixie AI is not available for your account's data region. You can connect your own AI."
Monthly allowance by plan
Every plan includes a monthly allowance of tokens for AI run by BI Pixie. Tokens are the units an AI service counts text in. The text BI Pixie sends and the text the AI writes both count, including the tokens the AI model uses to reason.
| Plan | Tokens a month |
|---|---|
| Free plan | 300,000 tokens |
| Standard plan | 1.5 million tokens |
| Pro plan | 6 million tokens |
| Scale plan | 6 million tokens |
| Enterprise plan | 25 million tokens, which BI Pixie support can raise on request |
The allowance belongs to your Microsoft organization (your Microsoft Entra tenant), not to one BI Pixie account. When several BI Pixie accounts in one organization use AI run by BI Pixie, they share one allowance, sized by the highest plan among them.
The allowance returns on the first day of each calendar month. The tokens are included in your plan and never billed. AI run by BI Pixie runs on Microsoft Azure OpenAI, not on your Fabric capacity.
See how much of the allowance is used
One meter shows how much of this month's allowance your organization has used, for example "1.1M of 1.5M tokens used", with the date the allowance returns. It appears in these places:
- In AI setup, on the Your AI row, when BI Pixie AI is your AI.
- On the AI Assistants page, on the Your AI for optimizations card, when BI Pixie AI is your AI.
- On the Account page, on the Usage tab, under Plan Usage. This row is shown whichever AI is your AI, so an administrator who switched to their own AI still sees the allowance waiting.
- In the BI Pixie AI option, when an administrator selects Change in AI setup.
Administrators and members both see the meter. Members see it read-only.
When 80% of the allowance is used, the meter turns to the warning color and adds the line "Your organization has used most of this month's AI allowance. New runs are refused when it is used up." No email is sent.
A run that started before the allowance ran out always finishes, so the used figure can go above the allowance, for example "1.6M of 1.5M tokens used". Nothing extra is charged.
When the allowance is used up
When the allowance is used up, a new run that needs your AI is refused before it starts, and nothing changes on the semantic model. The refusal says when the allowance returns and what the next plan includes, for example: "Your organization has used this month's AI allowance. It returns on November 1. The Standard plan includes 1.5M AI tokens a month."
An administrator sees two buttons: Set up your own AI, to keep going now, and an Upgrade button for the next plan. On the Pro and Scale plans the Upgrade button reads Get a quote. On the Enterprise plan there is no Upgrade button, and the refusal says that BI Pixie support can raise your organization's allowance. In the Fabric workload, Upgrade moves to the Plan tab inside your BI Pixie item, as the other Upgrade buttons do.
Members see the same refusal with no buttons, and it says that an administrator of your account can upgrade or connect your own AI.
When your organization's allowance comes from another BI Pixie account on a higher plan, upgrading your account would not raise it, so the refusal says which plan sets the allowance and offers no Upgrade button.
| Work | When the allowance is used up |
|---|---|
| Optimization previews: descriptions, AI instructions and display folders | Refused. |
| Writing benchmark questions | Refused. |
| Running a benchmark or an A/B test that BI Pixie would check with AI run by BI Pixie | Refused. The run is not started without its check. |
| The AI round of a data agent assessment | Refused. The assessment's other checks still run. |
| A semantic model assessment | Runs. The near-duplicate check is skipped, and the assessment says why. |
| A run that already started | Finishes. |
An optimization preview that would need more of the allowance than is left is refused before it starts, and you can choose fewer findings. Every other run starts as long as any allowance is left.
Switch to your own AI, or turn BI Pixie AI off
Only an administrator can change your AI. Members see "An administrator of your account can change your AI." on the Your AI row.
- To switch to your own AI, select Change on the Your AI row, choose the BI Pixie data agent, Azure AI Foundry, OpenAI or Anthropic, and save it. See Set Up Your AI for AI Readiness. Your own AI is never counted against the allowance.
- To switch back to BI Pixie AI, select Change, select BI Pixie AI, and then select Use BI Pixie AI. Your own AI stays saved under Generic AI agents to test, so you can switch back to it.
- To turn BI Pixie AI off, select Stop using BI Pixie AI on the Your AI row. If your account has an AI of its own saved, that AI becomes your AI again. If it has none, BI Pixie stops writing and grading with AI until an administrator chooses an AI again.
Once an administrator stops using BI Pixie AI, it never comes back by itself. A switch never stops a run that already started. That run finishes on the AI it started with.
What is sent to AI run by BI Pixie, and where
AI run by BI Pixie receives the same content your own AI would receive for the same work, and nothing more:
- When BI Pixie writes text for you, the part of the semantic model definition the task needs: names, data types, format strings, descriptions, DAX expressions, display folder names and the semantic model's AI instructions. When the reports built on the semantic model have Pixies, it also receives the titles people open and the values they filter to most, so that the text uses your own people's words.
- When BI Pixie phrases benchmark questions, the field names, the shape of each question and any filter value the question has to state. It never receives the correct answer at that stage.
- When BI Pixie checks a benchmark answer, the question, the recorded answer and the rows of the correct answer that BI Pixie calculated from your own data.
The only values from your business data that AI run by BI Pixie receives are the filter values and the rows of correct answers described above. It never receives telemetry event records or end-user identifiers, and it never receives your tables as a whole. Fields you excluded in the AI data schema contribute no values and no descriptions. When BI Pixie writes text, AI run by BI Pixie receives no values from a semantic model that defines row-level security, or from a column whose name indicates an identifier or personal data.
BI Pixie uses Microsoft Azure OpenAI under BI Pixie's own agreement with Microsoft, and processing stays in the geography of your account's data: the United States, or the EU Data Boundary. Microsoft's safety filter checks every request and response as it runs. Microsoft may keep the content for up to 30 days to look for abuse, and only Microsoft's authorized reviewers can see it. Microsoft does not use any of it to train AI models.
For the full text, see Privacy Policy Section 3.7 and the Data Processing Agreement.