The sources behind an answer became more visible

Microsoft’s October roundup described how Copilot would show the web searches derived from a user’s prompt in the citation section. Administrators also gained a separate policy for allowing or disabling web search independently of other connected experiences. Project teams therefore had a better basis for distinguishing organisational content from information retrieved from the web.

Loop could summarise longer documents, while Outlook could prepare meetings by collecting relevant context. These features reduced collection work but moved the review point: anyone using the summary needed to inspect which documents and searches were actually behind it, especially when the result informed a decision or was shared further.

Web-search control was also an architectural choice. If web content was required for market or project analysis, the team needed to document when it was used and what still came from internal sources. If web search was disabled, users needed to understand that an apparently current answer could be limited to the content Copilot could actually access.

Power BI showed what a report proposal used

The Power BI update on 15 October developed Copilot report creation further. Before creating a page, Copilot could ask follow-up questions and recommend fields and measures. The chat then displayed an outline of the content and data fields used. This gave the report author a more concrete basis for review than a finished chart on its own.

The update also announced that Copilot quick measure suggestions were being replaced by Fabric Copilot in DAX query view. Governing analytical work therefore had to cover feature changes as well as the quality of business definitions. A suggested measure could be syntactically valid while still using the wrong period, filter or business meaning.

This gave the analytics owner a concrete task: maintain a small catalogue of key measures with definitions, data sources, owners and approved uses. When Copilot suggested a field or measure, the author could compare it with the catalogue. A difference then became a visible quality issue rather than a hidden detail in a completed report page.

Collaboration did not guarantee a better result

A meta-analysis in Nature Human Behaviour, published on 28 October, examined when human–AI combinations produced better results. Its overall finding was that combinations did not systematically outperform the better party working alone, and outcomes varied with the task and interaction design. This challenged a simple assumption that inserting AI would always improve the work.

A project could use October’s features to establish a traceable process: retain questions and sources, inspect the report fields, and have a named subject specialist approve the conclusion. The useful target was fewer decisions with an unknown basis, rather than the highest number of Copilot actions. That distinction belonged in project goals, training and benefit measurement. The team also needed an update schedule, preventing old web evidence or an outdated report definition from being carried into new deliverables.

Sources

Microsoft Community Hub: What's New in Microsoft 365 Copilot | October 2024, 31 October 2024.

Microsoft Power BI Blog: Power BI October 2024 Feature Summary, 15 October 2024.

Nature Human Behaviour: When combinations of humans and AI are useful, 28 October 2024.

Forrester: With Agentic AI, Generative AI Is Evolving From Words To Actions, 8 August 2024.

For discussion

Which decision first needs a visible chain from question through source and data field to an approved conclusion?