Lab 10 — Researcher vs Analyst vs Chat: tool selection challenge

Level Duration Primary apps Sample files
200 20 minutes Copilot, Researcher agent, Analyst agent All datasets and documents

Department and industry focus

All departments and all six business units. Every role — HR, Legal, Marketing, Sales, Finance, Customer Service, Operations, and IT — faces the same choice between a quick answer, a researched briefing, and a calculated result, so this lab is deliberately industry-neutral and draws its examples from every other lab in the workbook.

Scenario

Choosing the right Copilot surface is a core skill. For each question, decide whether Copilot Chat, the Researcher agent, or the Analyst agent is the best tool, justify it, and define how you would validate the answer.

Learning objectives

Exercise — Decide and justify

# Question Best tool Why
1 Which departments are most over budget and by how much? Analyst Calculation and ranking over structured data with transparent logic.
2 How ready is Contoso Group for the QBR across all sources? Researcher Multi-source synthesis; needs attribution and evidence-vs-assumption split.
3 Rewrite this escalation update for an executive audience. Copilot Chat Single-step drafting from known context; no deep analysis.
4 Is CSAT worse for higher-priority tickets? Analyst Numeric comparison over structured data with method transparency.
5 What decisions should leadership make this quarter? Researcher Combines charter, risk, cost, and ops context into decisions.
6 Summarize the campaign brief into five bullets. Copilot Chat Quick single-source summarization; deep reasoning not required.

Forecasting and what-if questions — for example, what full-year variance looks like if spend holds at the current run rate, or what the quarter closes at if late-stage probabilities improve — belong with the Analyst agent or Copilot in Excel, because both can show the formula, assumptions, and columns behind a projected number. Copilot Chat is the wrong surface for them: it can describe a scenario in words, but it cannot calculate one reproducibly, and no surface should be asked to project beyond the periods the source data covers.

Reference guidance

Tool Best use Validation focus
Copilot Chat Quick summaries, drafting, single-source questions, brainstorming. Is the answer sufficient, or does it need deeper research or analysis?
Researcher Multi-step synthesis across files and broader context. Source coverage, evidence vs assumption, completeness.
Analyst Calculations, trends, correlations, code-backed data work. Columns, filters, joins, sample size, correlation vs causation.

💬 Debrief question: For question 5, when would you run Analyst first to produce evidence, then hand it to Researcher to synthesize? Discuss chaining agents.