Analysis became a dedicated agent conversation

Microsoft’s May roundup reported that Researcher and Analyst were beginning to roll out through the Frontier programme. Researcher focused on multi-step information work, while Analyst processed data. Copilot Notebooks also became generally available. A project team could collect sources, ask questions and develop a working basis in the same collaborative space.

The Power BI update on 19 May introduced a standalone Copilot experience called Ask Anything in preview. It was designed to answer questions across available data. Translytical task flows connected analysis with actions. The distinction between finding a pattern and changing a process therefore mattered: insight could be reviewed, while an action required authority and logging.

The data model set the boundary

Power BI also added an AI Data Schema for selecting which parts of a semantic model Copilot could use. This gave model owners a concrete way to reduce noise and guide language understanding. It did not repair weak definitions; a bounded model with the wrong customer concept or an unresolved measure could still produce a misleading answer.

Copilot Analytics gained a unified experience and new reports. Project leaders could see use, but needed to connect it with outcomes such as shorter analysis time, fewer errors or better decision papers. A plan should state which sources an agent used, who checked a calculation and when a conclusion required confirmation from a subject specialist.

Agent management in the admin centre exposed owners, data sources and actions, while consumption reports made costs visible. Unit cost became a useful measure: cost per reviewed analysis or completed task, rather than total AI consumption alone. An agent without an owner or active user group should be addressed before new agents were published.

Pressure to adopt was not evidence of impact

Microsoft’s Work Trend Index of 23 April drew on 31,000 workers in 31 countries and product signals. It reported that 81 percent of leaders expected moderate or extensive agent integration within 12–18 months. Because Microsoft supplied both the products and the research, this measured expectations rather than independently demonstrated benefits.

A May pilot should therefore begin with a question that could be checked against a known answer or expert assessment. The team could record sources, time, corrections and the resulting decision. If Researcher, Analyst or Ask Anything merely produced more answers without stronger control, the next step was to improve the data model and task definition before adding users.

Sources

Microsoft Community Hub: What's new in Microsoft 365 Copilot | May 2025, 20 May 2025.

Microsoft Power BI Blog: Power BI May 2025 Feature Summary, 19 May 2025.

Microsoft WorkLab: 2025 Work Trend Index Annual Report, 23 April 2025.

Gartner: Agentic AI in supply chain management solutions, 21 May 2025.

For discussion

Which analytical task has known sources, a verifiable answer and a named decision owner?