NEWS

Microsoft IQ: What Changes for Business Intelligence Agent Builders

What Microsoft IQ's November announcement means for BI agent builders: available features, previews, Power BI permissions and a practical pilot.

Safikul Islam
By Safikul Islam Published Nov 22, 2026 · 7 min read · 0 comments

Microsoft announced Microsoft IQ on November 17, 2026, bringing Work IQ, Fabric IQ, Foundry IQ and Web IQ together around reusable business context for AI agents. For a team choosing a business intelligence agent builder, the practical question is whether an agent can use the right business definitions and permitted data—not simply whether it can produce a fluent answer.

The Microsoft announcement includes a mixture of available capabilities, previews and future releases. It does not mean every IQ feature became generally available on announcement day.

Reviewed November 22, 2026. This article separates Microsoft’s published product status from suggested evaluation steps. It is not a hands-on deployment review.

What changed at Ignite—and what already existed?

The business implication is worth examining: teams may be able to reuse definitions and context instead of rebuilding them independently for every assistant.

However, Microsoft already documents a generally available Fabric capability in Microsoft 365 Agent Builder. That route connects an agent to accessible Power BI reports and semantic models for questions and analysis. It is not evidence that every newly announced Fabric integration is available.

Capability Published status at review Planning implication
Fabric capability in Microsoft 365 Agent Builder Generally available in current documentation Evaluate if you have the required licensing, permissions and tenant access
Fabric IQ in Excel’s Edit with Copilot; Fabric IQ connectors for ChatGPT and Claude Preview in the November 17 announcement Treat as evaluation candidates, not an assumed production dependency
Fabric data agents coming to Microsoft Copilot General availability described as coming soon Verify the exact integration and current release status before committing
Foundry IQ knowledge graphs and Microsoft Retriever 1 Public preview Test retrieval quality separately from your existing BI answers
Foundry IQ Serverless general availability Announced for December 2026 Future availability as of this article’s review date

Sources: the linked Microsoft IQ announcement, Agent Builder documentation, and Microsoft’s Foundry IQ release details. Read each status against the named capability; “Fabric” and “Foundry” are families of features.

Why the business definition matters more than the chat box

Microsoft’s Fabric IQ overview describes Power BI semantic models as the analytics layer for measures, dimensions and hierarchies. It also describes ontologies, currently in preview, as a way to represent business entities, relationships and rules.

For a business owner, a semantic model is useful because “revenue” needs a consistent definition. Does it include tax? Are refunds deducted? Is the date an order date or a payment date? A conversational interface does not settle those questions by itself.

Consider a hypothetical retailer with US and India operations. An owner asks, “Which region grew faster last month?” A credible answer needs a specified month, comparable revenue definitions and an explicit currency treatment. Adding USD and INR totals without conversion would produce a number that looks precise but answers nothing useful.

That is the evaluation opportunity behind this announcement: test whether business context survives the trip from the source model into the final answer. Do not use a good-looking demo as a substitute for this check.

The existing Agent Builder route: what to check first

Microsoft’s Agent Builder documentation lists Microsoft 365 Copilot Premium licensing for builders and users, plus permission and licensed access to the relevant Power BI content. The Fabric capability is read-only: it does not create, update or delete Power BI content. Sharing the agent does not grant access to the underlying reports.

The documented setup is to create or edit an agent, define its purpose in Configure, and enable Fabric under Work content. Some customers may see the label Power BI. Use Preview before sharing, then keep the initial audience small.

For a first BI assistant, I would define a narrow job: answer monthly sales questions from an approved model and explain the filters used. I would keep broader workflow automation out of that pilot so the team can judge the data answers clearly. For the general design process, see how to build an AI agent for your business.

US and India teams need to check processing location

The Fabric IQ in Copilot Chat documentation says Power BI data answering requires model processing in the United States or European Union. For Fabric tenants outside those geographies, an administrator must enable the relevant cross-geography processing setting. A team operating in India should check its actual tenant geography and organizational requirements before enabling it.

The same documentation says answers reflect the semantic model’s latest refresh, not automatically real-time data. It also lists unsupported content, including paginated reports, dashboards and top-level apps. This matters when choosing the report you intend to use for the pilot.

Existing access controls still matter. Test with the intended employee accounts, not only an administrator account that can see everything. If a regional manager should see one region, the acceptance test should include a request for another region’s results.

A practical acceptance test for a business intelligence agent builder

Build the test sheet before the demo. The following cases are suggested tests, not observed product results:

Question or condition What a useful answer should establish Reason to stop and investigate
“Compare net sales for October and September.” Correct measure, dates, filters and source A total that disagrees with the approved report
“Which region grew fastest?” Clarification of period and comparable currency basis Unsupported assumptions or mixed-currency arithmetic
The latest refresh failed An explicit statement about the available data period A stale result presented as current
A user asks for a restricted business unit Access remains limited to permitted data Disclosure beyond that user’s approved access
“Why did sales fall?” Separation of observed changes from possible explanations A confident causal explanation without evidence

Ask an analyst to prepare reference answers independently. Record correctness, missing context, response time and review effort. Repeat the same questions after changes to the model, permissions or agent instructions. A small, reproducible test set is more useful than collecting impressive screenshots.

Set the decision rule in advance. For example, require all permission checks to pass and all material numbers to reconcile before extending access. Let unresolved questions remain unresolved; an honest “I cannot establish that from this data” is preferable to an invented explanation.

How to judge the cost without guessing a price

There is no single verified all-in subscription price for the combined workflow in the sources reviewed here. Obtain a quote for your existing licences, additional users, required services and expected usage. Include the work of preparing the data and reviewing answers.

Use this planning formula: incremental monthly cost = new licences + metered services + maintenance and review time. Compare it with the value of the specific reporting work you expect to improve. Do not count already-owned licences as new spending, but do record whether they limit who can use the assistant.

Microsoft’s Foundry announcement describes upcoming Serverless consumption pricing and general availability in December. That is a roadmap item, not a verified November production entitlement. Keep a fallback reporting process if your plan depends on a future release.

Should a small business change platforms now?

If your team already maintains useful Power BI models, evaluate the documented Agent Builder route with a narrow internal pilot. If the definitions are inconsistent or the data refresh is unreliable, fix those problems first.

If your business uses another analytics platform, compare its existing agent capabilities before migrating. For example, GoodData announced Agent Builder on April 22, 2026, with configurable analytics agents connected to its governed data environment. That establishes another route to investigate, not a claim that the products are interchangeable.

My recommendation is to purchase against a specific reporting problem and a demonstrated workflow. Use the small-business AI tools guide to place this decision alongside your other software needs.

Frequently asked questions

Is Microsoft IQ a single new agent builder?

It is Microsoft’s broader context-management platform. Agent Builder is one experience to evaluate within that ecosystem. Check the named feature, licence and data source rather than treating IQ as one universal switch.

Can the Fabric capability update my Power BI data?

The documented Agent Builder capability is for read-only discovery and questions. Do not assume it gives the agent write access.

Do preview features mean the whole platform is in preview?

No. Status differs by capability. The table above separates the existing available route from previews and announced future releases.

What should I do this week?

Choose one approved report, write five business questions with known answers, confirm user access and licensing, and test privately. Decide whether the result reduces useful work before expanding the project.

Comments

No comments yet. Why don’t you start the discussion?

Leave a Reply

Your email address will not be published. Required fields are marked *