🔬
Researcher & Analyst · specialist agents
Choose research or data analysis
Researcher investigates questions across sources and produces a structured cited report. Analyst works across datasets to calculate, compare and explain patterns.
Business and enterprise access requires the Microsoft Copilot add-on and organizational enablement.
Start with the uncertainty you need to resolve
❓What is happening?Choose Researcher when the answer depends on gathering evidence.
📚Researcher investigatesIt compares work and public sources over several steps.
📊What is in the data?Choose Analyst when calculations and patterns drive the answer.
🧮Analyst tests the dataIt calculates, finds trends and explains exceptions.
Researcher resolves information uncertainty
🌐Investigate across sources
Combine public information with permitted work files, email, meetings and chats when the question spans both external and internal evidence.
⚖️Compare evidence
Evaluate vendors, competitors, strategic options or conflicting claims instead of asking Chat for a fast conclusion from limited context.
📑Produce a cited report
Researcher takes longer and returns an organized report with sources that can be opened, reviewed and shared.
Analyst resolves data uncertainty
🗃️Combine several datasets
Work across Excel files, CSV files and other supported data without manually consolidating every table before asking the business question.
📈Find trends and drivers
Compare performance over time, calculate statistics and test which regions, products or factors explain the movement.
🚨Surface unusual values
Identify outliers, missing patterns or exceptions that deserve investigation, then follow up to test possible explanations.
Pick the agent from the desired result
Choose
Question it answers
Typical result
Copilot Chat
Can I get a quick answer, summary or draft?
A fast conversational response.
Researcher
What does the available evidence support?
A deeper structured report with citations.
Analyst
What do these datasets calculate or reveal?
An analytical report with statistics, tables or charts.
Cowork
What actions and deliverables should follow?
Cross-app execution under your supervision.
Give each agent the right starting point
🔬Researcher prompt
State the decision, time period, source scope and standard of evidence. Ask it to show disagreements and missing information, not only a polished conclusion.
📊Analyst prompt
Attach understandable data and ask a focused question about a metric, change, driver or exception. Explain the business meaning of important columns.
Compare the three proposed customer-support platforms for our European business. Use public information from the last 12 months and the requirements in the attached brief. Evaluate security, integration effort, total cost and vendor stability. Show conflicting evidence and state where the available information is insufficient.
💡Key insightResearcher is not simply a slower Chat, and Analyst is not simply a spreadsheet summary. Use them when the work requires a deeper evidence process.
Deeper effort still needs human testing
🎯Frame the decisionMake clear what the work is meant to inform.
🔍Inspect the inputsCheck source quality or dataset structure before trusting depth.
🧪Test alternativesUse follow-ups to challenge the first explanation.
✅Verify the resultOpen citations and reproduce critical calculations.
Researcher can use different models and source scopes
🧠Model choice may vary
Researcher can expose GPT and Claude options when your organization enables them. Model choice can change reasoning style and output; it does not replace source review.
🌐Work, web or both
Choose the source boundary from the question. Internal strategy may need work context; market claims may need current public sources; a client brief may need both.
🗣️Clarification is useful
Researcher may ask follow-up questions before it begins. Answer them when scope, audience or evidence standards would materially change the report.
A citation is not a quality score
📅Check freshness and authority
A source may be old, promotional, secondary or outside the geography that matters. Open it before reusing an important claim.
🔗Check whether it supports the sentence
A cited page can be related to a claim without fully proving the specific conclusion that Researcher wrote.
Analyst needs a data-quality pass
Data issue
How it misleads
Useful follow-up
Missing values
Averages and trends may reflect only the records that remain.
Report missingness by field and segment before analysis.
Duplicate records
Volumes and totals can be counted more than once.
Identify likely duplicate keys and show their impact.
Changing categories
Year-to-year comparisons can combine unlike definitions.
Map category changes and rerun the comparison consistently.
Correlation only
A pattern may look like a cause when another factor explains both.
Test alternative explanations and state what the data cannot prove.
Use an iterative analysis conversation
Ask the business questionStart with the decision or outcome, not a request for every possible chart.
Inspect the first findingsCheck definitions, filters, missing values and unusual segments.
Challenge the explanationAsk which alternative factors could produce the same pattern.
Verify critical numbersReproduce important calculations before they drive a decision.
🛡️Important limitationMore processing can produce a more complete report, but it cannot repair weak sources or poor data automatically. Depth improves the method only when the inputs and conclusions are reviewed.