Researcher agent in Microsoft 365 Copilot
Gaurav Anand, CVP, Microsoft 365 Engineering Recent advancements in reasoning models are transforming chain-of-thought based iterative reasoning, enabling AI systems to distill vast amounts of data into well-founded conclusions. While some web-centric deep research tools have emerged, modern information workers need these models to reason across both enterprise and web data. For M365 users, producing thorough, accurate, and deeply contextualized research reports is crucial, as these reports can influence market-entry strategies, sales pitches, and R&D investments. Researcher addresses this gap by navigating and reasoning over enterprise data sources such as emails, chats, meeting recordings, documents, and ISV/LOB applications. Although these workflows are slower than the near real-time performance of Microsoft 365 Copilot Chat, the resulting depth and accuracy saves employees hours of time and effort. Our Approach Our approach mirrors the methodology a human would take when tasked with researching a subject: seek any needed clarification, devise a higher-order plan, and then break the problem into subtasks. They would then begin an iterative loop of Reason → Retrieve → Review for each subtask, collecting findings on a scratch pad until further research would unlikely yield any new information, at which point they would synthesize the final report. We instilled these behaviors into the Researcher with a structured, multi-phase process. Initial planning phase (P 0 ) The agent analyzes the user utterance and context to formulate a high-level plan. During this phase, the agent might ask the user clarifying questions to ensure the final output aligns with user expectations in both content and format. We define insights from this phase as I 0 . Iterative research phase The Researcher agent then loops through iterative cycles until it hits diminishing returns, starting with j = 1. Reasoning (R j ): Deep analysis to identify which subtask to tackle and what specific details are missing Retrieval (T j ): Search across documents, emails, messages, calendar, transcripts and/or web data to fetch the missing details. Review (V j ): Evaluating the collected evidence, computing its relevance to the original user utterance, and preserving the findings on a “scratch pad” We define ΔI j to be the new insights gained in iteration j from R j , T j , and V j . These are added to the prior knowledge: (I j = I j-1 ∪ ΔI j ). Note that with each cycle, the marginal insight ΔI j tends to diminish. The agent monitors this and essentially implements a check to conclude further research at iteration m when ΔI m < ε. Synthesis phase The agent synthesizes the aggregate I m by consolidating findings, analyzing patterns, drawing conclusions, and drafting a coherent report. The output includes explanations and cites sources to provide traceability. The Researcher agent in action To illustrate, if a user asks: "How did our Product P perform in Q4 compared to industry trends?”, the phases would be as follows. Planning Identifying subtasks: (1) get internal Q4 sales numbers for Product P; (2) find industry news or analyst reports on Q4 trends. Asking clarifying questions e.g., a specific region or competitor focus. Iterative research In iteration 1, it: Reasons: Start with internal sales data Retrieves: Pulls the Q4 sales report Reviews: Observes contribution of feature F in driving Product P’s sales growth In iteration 2, it: Reasons: Adapt the plan to explore feature F Retrieves: Retrieves internal and external communications about F; web search for competitor offerings Reviews: Customer reception of F; related industry news Iteration by iteration, it gathers pieces of the puzzle until new iterations yield only minor details. Synthesis Researcher then drafts a report, detailing a thorough comparison of Product P’s Q4 performance to the market, citing the internal sales numbers and external industry analysis, highlighting that feature F was a competitive differentiator. Technical Implementation Our current implementation leverages OpenAI’s deep research model, powered by a version of the upcoming OpenAI o3 model trained specifically for research tasks. Performance benchmarks highlight its efficacy, achieving 26.6% accuracy on the Humanity’s Last Exam (HLEx) and an average score of 72.6% on the GAIA reasoning benchmark 1 . Included below are a few technical approaches that were employed to build Researcher: Reasoning over enterprise data We have expanded the model’s toolkit with Copilot tools that can retrieve both first-party enterprise data—like meetings, events, and internal documents—as well as third-party content through graph connectors, such as shared company wikis and integrated CRM systems. These tools are part of the Copilot Control System that allows IT administrators and security professionals to secure, manage and analyze the use of Researcher. The Copilot tools are provided to the model using a familiar interface that the model was trained on such as the ability to “open” a document and “scroll” or “find” information within it. We have experimented with different techniques to address deviations from the distribution of the model’s original training data due to inherent differences between web and enterprise research queries. Internal evaluations revealed that the Researcher typically requires 30–50% more iterations to achieve equivalent coverage on enterprise-specific queries compared to its performance on public web data. Personalization with enterprise context Unlike web research where results are uniform regardless of user, Researcher produces highly personalized results. It leverages the enterprise knowledge graph to integrate user and organizational context, including details about people, projects, products, and the unique interplay of these entities within the user's work. For instance, when a user says, “Help me learn more about Olympus,” the system quickly identifies that Olympus is an internal AI initiative and understands that the user's team plans to take a dependency on it. This rich contextualization enables the system to: Ask more nuanced clarifying questions, such as: “Should we focus on the foundational research aspects of Olympus, or are you more interested in integration details?” Tailor the starting condition (P₀) for the deep research model so it’s not only precise but also personalized, thereby mitigating its lack of familiarity with company-specific jargon. Deep retrieval complementing deep reasoning Researcher retrieves a broad set of results for each query and semantic passages for each returned document to increase the insights gained per iteration T j. Instead of a serial iterative approach, Researcher first performs broad but shallow retrieval across heterogenous data sources and then lets the model decide the domains and entities to zoom into. Integrating specialized agents In enterprise contexts, interpreting data often demands the nuanced perspective of domain-specific experts. That’s why agents are a critical part of the Microsoft 365 Copilot ecosystem. Researcher is being extended to seamlessly integrate with other Agents. For instance, Researcher can leverage the Sales Agent to apply advanced time-series modeling to provide an insight like, “Sales in Europe are expected to be 5% above quota, driven by product X,” Moreover, these tools and Agents can be chained together. For example, if a user asks, {help me prepare for my customer meetings next week}, the system first employs calendar search to identify the relevant customers; and then, in addition to pulling searching over recent communications, it also retrieves the CRM information from the Sales agent. By allowing Researcher to delegate complex subtasks to these specialists, we help compress multi-step reasoning iterations into a single step and complement Researcher agent’s intelligence with specialist knowledge. Results and Impact Even in early testing, Researcher has demonstrated tangible benefits. Response quality We evaluated Researcher extensively in early trials, focusing on complex prompts that require consulting multiple sources. For quality assessment, we employed a framework called ACRU, which rates each answer on four dimensions: Accuracy(factual correctness) Completeness (coverage of all key points) Relevance(focus on the user’s query without extraneous info) Usefulness(utility of the answer for accomplishing the task) Each dimension is scored from 1 (very poor) to 5 (excellent) by both human and LLM-based reviewers. When we compared Researcher’s performance against our baseline M365 Copilot Chat on a diverse set of 1K queries, we saw an increase of 88.5% in accuracy, 70.4% increase in completeness, 25.9% increase in relevance, and 22.2% increase in utility. It is worth noting that the agent’s improved accuracy comes from its ability to double-check facts. It cites on average ~10.1 sources per response in our above evaluation. 61.5% of the answers included at least one enterprise document as a source, 58.5% included a web page, 55.4% cited an email, and 33.8% pulled in a snippet from a meeting transcript. Time savings For this measurement, we surveyed two groups of internal users: 22 Product Managers responsible for crafting product strategy documents and project updates to align stakeholders 12 Account Managers interacting with Microsoft customers, writing client proposals, and maintaining clear communication with stakeholders The feedback from both groups has been extremely positive. Users reported tasks that previously took days of manual research could be completed in minutes with the agent’s help. Overall, our pilot users estimated that Researcher saved them 6–8 hours per week, essentially eliminating an entire day’s worth of drudgery. Here is verbatim from a product manager “it even found data in an archive I wouldn’t have checked. Knowing the AI searched everywhere—my meeting transcripts, shared files, the web—makes me trust the final recommendation much more.”. I have found myself using Researcher daily. Researcher’s intelligence to reason and connect the dots leads to magical moments. Below is a snippet from a report to prepare for my upcoming meetings. The appointment at 11:30am was a placeholder for me to send out broad communication to the team with some survey results. Researcher identified that I had done this already and encouraged me to use the time instead to collect feedback from the team. What's Next Reinforcement Learning We will continue to improve the quality of Researcher to make reports more complete, accurate and useful. The next phase of adaptation to enterprise data will involve post-training reasoning models on real-world, multi-step work tasks using reinforcement learning. This will involve learning a policy function (π(s)→a), which picks the next step a as a function of its current state s to maximize the cumulative reward: Steps are range of actions accessible to the model (reasoning, tools, synthesizing) State encapsulates the user’s initial utterance and the insights I n thus far Reward function evaluates output quality at each decision point Formally, we interleave internal reasoning and actions to build the cumulative insight I(i)=I(i-1)+R(s i ,a i ), where (R(s i ,a i )) denotes the reward obtained by taking action a, given the state s i. Through successive iterations, the model learns an optimized policy ((π(s)). To achieve this, we will focus on creating datasets for high quality research reports and investing in robust evaluation metrics and benchmarks. User control Researcher reasons across knowledge sources that the user has access to and find the most useful nuggets of information. However, we understand our users and enterprises often need more control over the information sources. To this end, Researcher will allow “steerability” over the sources from which the report will be created. Below is an early visual of what this could look like. Agentic orchestration Agentic orchestration is a core capability of Researcher. We have already integrated a few Microsoft agents, and we will generalize this capability. Moreover, we will afford end users and admins the ability to customize Researcher by bringing their own agents into the Researcher workflow. For example, imagine a law firm has created an agent to format reports into legal briefs. We will allow the output of Researcher to be chained with this custom agent to customize the output. Conclusion Researcher can significantly transform knowledge workers’ everyday tasks. Early results show that users trust the agent to deliver factually accurate and detailed reports that save time and drive productivity. As we expand the capabilities of Researcher, improve quality and allow deeper customization, we envision a future where Researcher evolves into a trusted and indispensable tool in the workplace. For additional details on Researcher, including rollout and availability for customers, please also check out our blog post highlighting reasoning agents within M365 Copilot and more. 1 Introducing deep research | OpenAIIntroducing Copilot Memory: A More Productive and Personalized AI for the Way You Work
At Microsoft, we believe AI should work for you—not the other way around. That’s why we’re introducing memory in Copilot, a new capability that makes your Copilot more productive, more helpful, and more in tune with how you work.Analyst agent in Microsoft 365 Copilot
Xia Song, CVP, Microsoft 365 Engineering As large language models (LLMs) and multimodal systems revolutionize information work by seamlessly navigating language, code, vision, and voice, a vast domain of structured, tabular data remains underutilized: Excel sheets, databases, CSV files, and Power BI reports often lack the natural intuitiveness of text or images. Picture a project manager urgently seeking quarterly performance insights scattered across multiple Excel worksheets and a badly formatted table inside a presentation. Some metrics are hidden in the middle of a worksheet, while some TSV files use commas instead of tabs—leaving little guidance on which data matters or how it connects. For those unskilled in data wrangling, this scenario can devolve into hours of frustration or missed insights. Yet armed with the know-how to manipulate data and harness code as a tool, one can swiftly unravel such complexity, extracting pivotal information and gaining a critical competitive edge. But what if everyone had this capability readily available? That’s precisely the motivation behind the launch of Analyst, one of the first reasoning agents of its kind in M365 Copilot. Powered by our advanced reasoning model, post-trained on OpenAI o3-mini on analytic tasks, Analyst acts as your “virtual data scientist”. This reasoning-powered agent is built directly into Microsoft 365, placing sophisticated data analytics capabilities right at your fingertips. The Era of Progressive Reasoning and Problem Solving Traditional LLMs have historically jumped too quickly from problem to proposed solution while often failing to adjust to new complexities or gracefully recover from mistakes. The advanced reasoning model behind Analyst changes this by implementing a reasoning-driven, chain-of-thought (CoT) architecture derived from OpenAI’s o3-mini. Instead of providing quick answers, it progresses through problems iteratively by hypothesizing, testing, refining, and adapting. Analyst takes as many steps as necessary, adjusts to each complexity it encounters, and mirrors human analytical thinking. With the capability to generate and execute code at every step within its reasoning trajectory, this model excels at incremental information gathering, constructing hypotheses, course correction, and automatic recovery from errors. Real-World Data is Messy: A Case Study Real-world data is messy. To illustrate the tangible benefits of the model’s reasoning capabilities, let's consider a practical challenge. Imagine you're presented with two datasets: Dataset A: An Excel file with multiple sheets containing data on world internet usage, where the critical data isn’t conveniently located at the top left but located somewhere in the middle of the second sheet. Dataset B: A .tsv file containing data on country statistics, presumably tab-delimited, but mis-formatted with commas as delimiters due to an export error. The task at hand? Vague at best—something like, “Help identify and visualize interesting insights and connections between these two datasets”. Most of the traditional tools and existing models struggle here. They either stall entirely or deliver incomplete or incorrect analyses. However, when faced with precisely this scenario, Analyst demonstrates remarkable resilience: It quickly identifies and navigates directly to relevant data hidden in the middle of an Excel sheet. Shows curiosity, discovers then lists the sheet names. Gracefully detects and corrects delimiter issues in the second dataset. Progressively explores the data through iterative hypothesis-testing steps, constructing actionable insights without explicit guidance. As a result of the progressive problem solving shown, the model handles these complexities smoothly and provides observations, insights and visualizations all by itself, demonstrating transformative potential in real-world analytic tasks. How It Learns: Reinforcement Learning, Structured Reasoning, and Dynamic Code Execution The effectiveness of the advanced reasoning model behind Analyst lies largely in reinforcement learning (RL). Built by post-training OpenAI’s o3-mini model, it employs advanced RL coupled with rule-based rewards to handle extensive reasoning paths, incremental information discovery, and dynamic code execution. We’ve observed that model performance consistently improves with more reinforcement learning compute during training, as well as more deliberate thinking during inference. Analyst takes advantage of STEM and analytical reasoning optimizations introduced by models like o3-mini, excelling in structured data scenarios. It dynamically writes, executes, and verifies Python code within a controlled execution environment. This iterative cycle enables the model to continually adjust its strategy through course corrections and effective recovery from errors, emulating human problem-solving behavior closely. Data Diversity and Robust Reward Training data diversity is fundamental to post-training effectiveness. We built extensive datasets that encompass a wide range of real-world enterprise scenarios and structured data types: File types: Excel, CSV, TSV, JSON, JSONL, XML, SQLite databases, PowerPoint presentations, etc. Similarly, the task variety ranged from straightforward numerical computations and visualizations to exploratory hypotheses construction and prediction. The data points used in training were carefully constructed and selected to represent authentic complexity, preventing our model from overfitting any particular task or benchmark. Recognizing the "reward hacking" behavior often observed in reinforcement learning systems that can potentially lead to model capability loss, we refined our reward systems via adopting more advanced and robust graders. This meticulous data selection, combined with rigorous task design, was done to ensure genuine reasoning by incentivizing authentic exploration and accurate outcomes. Results The following benchmark results further underscore our model’s strengths on rigorous analytics-focused tasks like those in DABstep benchmarks and internal M365 Copilot comparisons. DABStep (Data Agent Benchmark for Multi-step Reasoning) DABStep is a rigorous evaluation suite designed to test AI agents on real-world data analysis and reasoning tasks. It consists of 450+ structured and unstructured tasks, split into Easy and Hard categories. The Easy set involves simpler data extraction and aggregation, while the Hard set requires multi-step reasoning, integration of diverse datasets, and domain knowledge. When benchmarked against DABStep, our model demonstrated overall state-of-the-art performance among known baselines. It showed excellent capability on both simple and complex tasks, with a substantial lead in the latter category. Note: The M365 Copilot Analyst Model currently appears in the real-time leaderboard as an unvalidated anonymous submission labeled "Test1". We have contacted the DABStep team to update the submission and reflect this as the Analyst model coming from Microsoft. Product Benchmarks While academic benchmarks provide valuable insights, the true measure of a model’s value lies in its practical application within real-world scenarios. We benchmark our model’s performance on enterprise data analysis tasks across diverse business documents, including Excel spreadsheets, CSVs, PDFs, XMLs, and PowerPoint files, reflecting common analytical workflows within the M365 suite. We compare the specialized Analyst agent against the mainline M365 Copilot Chat (without deep reasoning), evaluating their accuracy in insight generation, data interpretation, and structured query execution across various enterprise file formats. Analyst is powered by the advanced reasoning model which consistently outperforms existing approaches, demonstrating not only incremental but transformative improvements in real-world analytic reasoning. The Road Ahead: Opportunities and Acknowledgements We are genuinely excited about what Analyst can unlock for Microsoft 365 users in making advanced data analytics capabilities accessible to every user. Yet we remain conscious of current limitations, recognizing plenty of room for further improvement. Opportunities remain for more seamless integration across applications, improved interaction paradigms, and expanded model capabilities to handle an even broader spectrum of analytic scenarios. We are committed to continuous improvement of Analyst and the underlying model, listening closely to user feedback, and refining our model and its integration with other products. Our ultimate goal remains clear: to empower users and organizations to achieve more, turning everyday information workers into empowered analysts with a “virtual data scientist” at their fingertips. For additional details on Analyst, including rollout and availability for customers, please also check out our blog post highlighting reasoning agents within M365 Copilot and more. References: Introducing Researcher and Analyst in Microsoft 365 Copilot OpenAI o3-mini Learning to reason with LLMs DABStep: Data Agent Benchmark for Multi-step ReasoningThe Microsoft 365 Copilot app: Built for the new way of working
The Microsoft 365 Copilot app is designed for the new way of working. Copilot is woven into your daily workflow—offering tools like secure AI chat, agents, content creation, and AI-powered enterprise search—all in one place. Learn how features like Copilot Notebooks, the Agent Store, and personalized memory can help you move quickly, focus on the content that matters across Windows, web, mobile, and soon macOS.7 great Copilot prompts for the holiday season
This holiday season, Microsoft 365 Copilot is here to help you and your organization free up time and focus on what’s important. We’ve put together seven of our favorite Copilot prompts for preparing for out of office, planning holiday events, and catching up on work quickly. Share the prompt examples below with your users to inspire greater Copilot usage during the season! 1. Thank your team (Outlook) A heartfelt end-of-the-year message can make your team feel valued and recognized for their hard work, setting a positive tone for the upcoming year. Prompt: Write an end-of-the-year message to my team congratulating them on a great year and all the work on [Project Monaco]. Use a warm and appreciative tone with some light humour. Consider incorporating bullet points to highlight specific achievements or milestones. Try it in Outlook – Create a new email, select “Draft with Copilot” and paste this prompt. Pro tip: Use Copilot to tailor the email further, like adjusting the tone or length of the email or including the names of your co-workers in the prompt to make the response more personal. 2. Set your out-of-office message (Business Chat) Need to setup an out-of-office reply? Copilot can draft a festive message for you, spreading holiday cheer while ensuring your contacts know when you'll be back and who to reach out to in your absence. You can even request a reminder for how to set Out of Office replies in Outlook. Prompt: Write some funny Out of Office email responses to use while I'm on vacation from [Dates]. Also include steps for how to set this in Outlook. Try it in Business Chat via this link. Pro tip: Ask Copilot to turn the messages into poems! 3. Keep your team updated on important holiday details (Word) Be confident everyone is on the same page regarding office hours, vacation schedules, and key deadlines. Leverage this prompt to create a template that you can customize and share with your team to maintain smooth operations and foster a supportive, well-coordinated team environment. Prompt: Create a template for a holiday schedule that outlines office hours, team members' vacation days, and any important deadlines during the holiday season Try it: Create a blank document on Word, and ask Copilot to generate this template for you. Pro tip: Select the paperclip icon to attach relevant emails, files, or meetings as references to the prompt to further tailor Copilot’s response. 4. Kickstart team event planning (Business Chat) Need help with catering for a team event? Streamline the party planning process with Copilot. From menu comparison to cost analysis, Copilot will help with the research and ensure you make the best choices for your team. Prompt: List options for a local catering company that can support a holiday event for 20 people. Be sure to include companies that have vegetarian and gluten-free options. Try it in Business Chat via this link, or paste the above prompt into Copilot. Pro tip: Adjust the prompt and ask Copilot to look for catering companies within your city or town and have options for delivery. 5. Brainstorm gift options (Business Chat) Copilot can be your go-to assistant for brainstorming gift ideas this holiday season. Use Copilot to help you find the perfect gift to make your colleagues, friends, and family feel special. Prompt: Suggest some holiday gift ideas for my clients that are under $50 each. Ensure the list contains innovative and engaging items that ensures these are memorable. Please include a brief description of what the item is, how much it costs, and the expected reaction from different client personalities. Make the title of each item in the list a link of where to purchase it. Try it in Business Chat via this link, or paste the above prompt into Copilot. Pro tip: Experiment with this prompt to get a more personalized response – make sure the toggle is set to Work mode and ask Copilot to suggest gifts for the top 5 people you’ve recently collaborated with. 6. Catching up after the holidays (Outlook) Let Copilot help you get up to speed on important emails after you have taken some well-deserved time off. Prompt: Summarize the emails I've received over the past two weeks and prioritize them based on urgency. Try it in Outlook by selecting the Copilot icon on the right side of the top banner in the new Outlook application (or Outlook on the web). 7. Prepare for your first day (Business Chat) Feel confident on your first day back to work after the holidays by using Copilot to prepare for your meetings. Copilot will provide a recap of relevant meetings and communications. Prompt: Help me prepare for my next meeting. Try it in Business Chat via this link, or paste the above prompt into Copilot and make sure it’s toggled to Work. As you celebrate the season, let Microsoft 365 Copilot be your organization’s holiday helper. Combining specific, useful prompts with the right Microsoft 365 applications ensures users are making the most out of Copilot. For additional Copilot prompt inspiration and skilling, encourage people at your organization to explore the Copilot Prompt Gallery, where they can discover, save, and share their favorite prompts. Looking for more resources to drive Microsoft 365 Copilot adoption? Check out new Copilot user engagement tools and templates on the Microsoft Adoption site, including: User onboarding email templates (download link) Manager onboarding resources (download link) Happy Holidays! Please note that outputs may vary when using these Copilot prompts. Large language models output varying responses, even for the same exact prompts. When utilizing these suggested prompts, expect new/fresh answers from Copilot. Users should still review the outputs from Copilot to ensure that the final output aligns with the users’ goals.
