Microsoft Adds ‘Deep Reasoning’ to Copilot AI for Data Analysis

Microsoft’s AI chatbot software, Microsoft 365 Copilot, is launching a “deep research” AI-powered tool. Following Tuesday’s AI news from Google and OpenAI, Microsoft has made its own announcements. These include information about two “deep reasoning” agents for Microsoft 365 Copilot, named Researcher and Analyst, which the company claims are the first of their kind, as well as new features for custom AI agents. In order to do “complex, multi-step research,” the researcher uses OpenAI’s deep research AI model. Additionally, third-party data is accessed through connections to sources such as Salesforce or ServiceNow, allowing business clients to obtain insights from all of their tools.

Microsoft on Tuesday launched two sophisticated AI agents — Researcher and Analyst — meant to add deeper reasoning skills to Microsoft 365 Copilot. The capabilities will begin rolling out in April through the company’s new early access program to select users.

Numerous deep research agents have been introduced recently for chatbots, such as xAI’s Grok, Google’s Gemini, and OpenAI’s ChatGPT. They are powered by reasoning AI models, which have the capacity to solve issues and verify their own information, two abilities that are undoubtedly crucial for carrying out in-depth research on a topic.

Microsoft claims that with chain-of-thought reasoning, Analyst, which is based on OpenAI’s o3-mini reasoning model, can create spreadsheets from raw data, run Python code that you can see while it runs, and function at the level of a proficient data scientist, creating reports like this one (pdf).

Along with additional autonomous agent features that are beginning to roll out in Copilot Studio, those tools are set to begin rolling out in April to Microsoft 365 Copilot license users as part of an early access program.

The varieties offered by Microsoft are referred to as Analyst and Researcher. By securely combining work data, such as emails, meetings, files, and chats, with external sources throughout the Web, the new agents are intended to offer in-depth, expert-level support.

To assist users in completing challenging research tasks, Researcher integrates Microsoft 365 Copilot’s orchestration and search features with OpenAI’s research model. According to Microsoft, the agent is designed to provide insights of a better caliber and more accuracy than were previously achievable.

“You can use Researcher to build a detailed go-to-market strategy based on the context of all your work data and broader competitive data from the web; identify whitespace opportunities for a new product based on emerging trends and internal data; or create a comprehensive quarterly report for a client review detailing work to date along with the latest market analysis,” Microsoft’s Jared Spataro wrote in a blog post.

With rule-based workflows that incorporate AI activities, Microsoft asserts that the new agent flows in Copilot are strong enough to “automate any task you can imagine.” We’ll have to see it in action to see how it’s better than adding a checkbox or two, how well its “low code” experience actually works, and whether it can live up to the promises AI companies are making about agents. The LinkedIn announcement describes scenarios like an agent flow that routes feedback emails to the appropriate team.

Researchers integrate “advanced orchestration” and “deep search capabilities” with OpenAI’s deep research model, which drives the company’s proprietary ChatGPT deep research platform. According to Microsoft, a researcher can do analysis, such as formulating a go-to-market plan and preparing a client’s quarterly report.

Additionally, Researcher facilitates connectivity with third-party platforms like Confluence, ServiceNow, and Salesforce.

According to Microsoft, Analyst is “optimized to do advanced data analysis” and is based on OpenAI’s o3-mini reasoning model. Iteratively, the analyst works through issues, improving its “thinking” and offering thorough responses to questions. According to Microsoft, analysts may also use Python, a programming language, to solve intricate data queries and make their “work” available for review.

The ability to access work data and the global web is what gives Microsoft’s deep research tools a modest edge over the competitors. For instance, researchers may access data from AI “agents,” tools, and applications such as Confluence, ServiceNow, and Salesforce by utilizing third-party data connections.

Analyst, a data analysis tool driven by OpenAI’s o3-mini reasoning model, was also introduced. The business claims that the bot simulates human analytical thinking by using Python-based computing and chain-of-thought reasoning.

The true problem, of course, is making sure that programs like Researcher and Analyst don’t make things up or have hallucinations. Deep research and models, such as o3-mini, are far from flawless; occasionally, they miscite previous research, make erroneous inferences, and use questionable public sources to support their arguments.

Researchers and analysts will be available to Microsoft 365 Copilot clients through a new Frontier initiative that Microsoft is establishing. Beginning in April, those registered in Frontier—which will first get experimental Copilot capabilities—will receive Researcher and Analyst features.

The program demonstrates Microsoft’s ongoing efforts to integrate reasoning skills into the workplace, providing “expertise on demand” as it relates to day-to-day tasks.

Platform for Agent Deployment in Copilot Studio Additionally, Microsoft launched Copilot Studio’s deep reasoning and agent flows, which let businesses create agents that can automate operations using predictable patterns or carry out intricate, multi-step business processes.

In addition to the new platform, Copilot Studio has launched additional capabilities that will enable autonomous agents to operate on their own, initiate events, and oversee business operations without human assistance.

Microsoft stressed that the Copilot Control System, which provides enterprise-level data grounding, access governance, and compliance controls for IT teams, will support all of the announcements.