New reasoning agents: Researcher and Analyst in Microsoft 365 Copilot

Analyze data and research with expertise on demand, and automate workflows with intelligent agents in Microsoft 365 Copilot. Analyst thinks like a data scientist and Researcher like an expert, so you can uncover insights, validate logic, and generate expert-level reports in minutes.

And using Microsoft Copilot Studio, build your own autonomous AI agents to streamline multi-step processes with deep reasoning, like responding to RFPs or synthesizing internal knowledge, incorporating Copilot Flows as automated actions. No need for perfect prompts—just describe what you need, and Copilot will reason through the task, surface key insights, and deliver actionable results faster than ever. 

Jeremy Chapman, Microsoft 365 Director, walks you through how to use these AI-driven agents step-by-step, from data analysis to research and automation.

► QUICK LINKS:

00:00 - Reasoning agents in Microsoft 365 Copilot

00:59 - Analyst 02:58 - Researcher

04:46 - How it differs from other models

05:29 - How to build your own deep reasoning agent

07:58 - Wrap up

► Link References

To get started, check out https://aka.ms/CopilotReasoning

► Unfamiliar with Microsoft Mechanics?

As Microsoft's official video series for IT, you can watch and share valuable content and demos of current and upcoming tech from the people who build it at Microsoft.

• Subscribe to our YouTube: https://www.youtube.com/c/MicrosoftMechanicsSeries

• Talk with other IT Pros, join us on the Microsoft Tech Community: https://techcommunity.microsoft.com/t5/microsoft-mechanics-blog/bg-p/MicrosoftMechanicsBlog

• Watch or listen from anywhere, subscribe to our podcast: https://microsoftmechanics.libsyn.com/podcast

► Keep getting this insider knowledge, join us on social:

• Follow us on Twitter: https://twitter.com/MSFTMechanics

• Share knowledge on LinkedIn: https://www.linkedin.com/company/microsoft-mechanics/

• Enjoy us on Instagram: https://www.instagram.com/msftmechanics/

• Loosen up with us on TikTok: https://www.tiktok.com/@msftmechanics

 

Episode Transcript

(00:02):
Chain-of-thought reasoning
is coming to Microsoft 365 Copilot.
Today, I'll show you an early look
at two new reasoning agents for work,
as well as how you can buildyour own deep reasoning agents
in Microsoft Copilot Studio.
Starting with Analyst,
which uses reasoning to work
alongside you as a data scientist,
helping you to go from rawdata to valuable insights

(00:23):
in minutes using analytical reasoning
while exposing its underlying query code.
Then Researcher that workswith internal information
that you can access, togetherwith powerful orchestration
and deep search to tackle complex,
multi-step research topics.
Now, these agents bothuse OpenAI's o3 model
and can collaborate with you,

(00:44):
prompting you along the way
like a coworker to generateadvanced responses.
Both are designed to analyze large amounts
of work information that youhave permission to access
to deliver on-demand expertise
so that you can get more done.
Let me show you Analyst first.
So we built Analyst to thinklike a skilled data scientist.
Now, Analyst leverages astate-of-the-art reasoning model

(01:07):
that we've optimized foradvanced data analysis at work.
In this case, I'm trying to understand
our most loyal customers.
Now, here I have this messy dataset.
It's spread across multiple sheets,
and in each, there are thousands of rows
of information across customers.
Now, here you can see the monthlyrevenue for each customer,
and none of this has beencleaned or contextualized.

(01:30):
And now I have an analytics expert
available to me 24/7 that can help.
Notice I don't have to spend time
writing the perfect promptto get what I'm looking for.
I only need to ask Copilot to use
the data that I've referencedto give me insights
on customer segments witha graph to visualize it.
Now, you'll see theagent takes my question

(01:51):
and you can watch itsreasoning as it processes.
So first, it's analyzing the data
to look for the right columnsfor customer segmentation
and revenue over time.
Now it's mapping out clustersand also grouping customers.
It's getting an understandingthen of the sheet structure
in my Excel file, standardizing the data,
loading what it needs from the data,

(02:12):
and then prepping it for further analysis.
Now, it starts to workon data visualization,
looking for patterns and alsorevenue trends in this case.
It ensures that the data is complete,
and while it's working,
you can also click in to expand
any of these steps to see itschain-of-thought reasoning,
as well as the Python codethat it's running in real time.

(02:32):
Now, that way, I can also validate
the approach that it's taking.
Now, as Analyst runs its finalstep and finds key insights,
it starts working on thevisualization that I asked for.
Now it's created an AverageMonthly Recurring Revenue
by Customer Segment chart
based on customer size and the state
of their monthly revenue over time.
Below that, there's even a summary
of its findings pointingout three key insights

(02:55):
from the data that are easy to understand.
Next, with Researcher,
we're using OpenAI's deep research model,
along with advancedorchestration in Copilot
and vector-based search over the work data
that you have access to,
to deliver complex, multi-stepresearch with high accuracy.
Now, in this case,
I work in productdevelopment and my company

(03:15):
is actually entering a new market,
so I need help developing
a product strategy for our expansion.
Now, I just need to write
A Prompt Here And I'llAsk Researcher to develop
a product strategy to enter
a new market segment foroutdoor and adventure goods.
Once I enter my prompt,Researcher goes to work.
You can see that as partof its first response,

(03:36):
it asks me clarifying questions
about the scope and formatof what I want to write.
So I'm going to go ahead
and respond to Researcher,answering both of its questions.
And then it uses myresponse to move forward.
The agent takes my prompts,understands the task,
and starts to construct a plan of action
that it will use toauthor a detailed report.

(03:56):
Then it starts reasoningover multiple files
that I can access frominternal information sources.
As it works, I can take a look
at its chain-of-thoughtreasoning in real time,
and it tells me what it's doing,
the topics it's searching for,
and the files and messages it's using
as it completes the task in real time.
Now it's building an understanding
of my existing product lineup,referencing recent meetings,

(04:20):
and also even researchingindustry trends from the web.
Now, this process takes a few minutes,
so let's go ahead andjump ahead to the result.
On the right side,
you can see that it's delivered
a thorough response withfully-documented product strategy,
in line with what I'dexpect from an expert.
Starting with an analysisof my existing business,
it's also analyzed the Outdoorand Venture Gear Sector,

(04:43):
then built insightsbased on the intersection
of our existing business, electronics,
with our new Outdoor product segment,
then Strategic Positioning,
and also a detailed Go-to-Market plan.
And this goes beyond
what you can do with other models
that are designed to reason
primarily over contentsourced from the web.
So in this case,

(05:03):
Researcher leveraged internal work files
and information that I have access to
to get the most recent andrelevant data from files,
email messages and meetings,
in addition to informationthat it sourced from the web.
And Researcher can even beconnected to third-party data
via connectors like Salesforce,ServiceNow, and Confluence,

(05:24):
or even other agents likeSales Chat from Microsoft.
Analyst and Researcher are, of course,
prebuilt agents that you canuse for Microsoft 365 Copilot.
But now, why don't we switch gears
to how you can add deepreasoning to the agents
that you build yourself usingMicrosoft Copilot Studio.
Now, in this case,
my team frequently receivesrequests for proposals,

(05:46):
or RFPs, from its clients,
and we'll use Copilot Studio,which, if you're new to it,
is a platform to create,manage, and deploy agents.
In this case, we canuse it to build an agent
that works independently to help automate
the authoring of the initial proposal.
Now, the thing about RFPs
is that they often needextensive collaboration

(06:07):
with people who also need access
to the right informationfor pricing and scheduling,
and the requesters, they expect detailed
and accurate proposalsto win their business.
Now, agents in Copilot Studiocan help your team complete
an ambiguous or multi-partprocess like writing an RFP,
and you can use deep reasoning
in Copilot Studio to instruct

(06:27):
an agent to create aformal proposal document
based on that RFP.
All you need to do isprovide detailed instructions
that break down the stepsthat you want it to take.
For example, here, I've instructed
the agent to use CRM data
and product informationstored across various files
in SharePoint to find answersto our client's questions.
Next, I'm going to instruct

(06:48):
the agent to use reasoningmodels for the process,
and then moving into settings,
I've already connected theagent to my knowledge sources,
in this case, files in SharePoint,
and I've also selected Actions
which are connections to my CRM system
and a Flow that I've builtto create the proposal.
I've also added a trigger
to kick off the processwhen a new RFP email arrives

(07:09):
in Outlook to make thisan autonomous agent.
And with everything configured,
I can even test it using an RFP
that was successfully processed previously
to validate it's running correctly.
And I'm going to go aheadand choose this one here
from 3:00 PM and now I canwatch the agent's process
in realtime as it runs the steps
to author the proposal document.

(07:29):
The agent is capable ofquickly analyzing large amounts
of data to generate precise
and well-thought-outresponses and delivers
a well-formatted RFP.
And when I publish the changesI've made to the agent,
it will become active.
So now, once an RFP arrives in my inbox,
the autonomous agent wejust built will process
the RFP to generate amatching proposal document,

(07:52):
then send that as afirst draft to the team
so that we can finalizeit and respond faster.
So using this approach,
you can build agents in minutes
to complete processes for you
using built-in deepreasoning capabilities.
And to find out more about agents
with deep reasoning availablewith Microsoft 365 Copilot
and your options for using Copilot Studio,

(08:12):
check out aka.ms/CopilotReasoning.
And keep watching Mechanics
for the latest updates from Microsoft,
and we'll see you soon.