Your company has provided Microsoft 365 Copilot. You have tried rewriting emails and summarising documents, but performance analysis or a management presentation may still begin with a vague request: “Analyse this for me.”
For professionals expected to work independently, coordinate colleagues and make recommendations, the next step is to give Copilot a defined place in the analysis: check the evidence, compare options, then prepare the recommendation. These four prompts from iTalent take you from the first question to a decision brief you can verify.
To try the examples, paste the sample material and prompt into your organisation’s Copilot chat, then work through the follow-up questions. For actual documents, identify the version to use and check that the response cites it. File and work-data access depend on your account setup; these exercises can begin with pasted text.
Start with a decision brief
Instead of uploading a folder and asking for analysis, define a choice: “Should we fix follow-up next month or increase campaign spending?” Microsoft’s prompting guide recommends a goal, context, sources and expectations. For analytical work, add constraints and evidence requirements.
Decision: [specific choice, not just a topic].
Reader and use: [decision-maker and meeting].
Success measure: [metric, definition and period].
Hard constraints: [budget, capacity, deadline and non-negotiable requirements].
Supplied evidence: [file names, versions, periods and field definitions].
Unknowns: [missing data or unconfirmed arrangements].
Separate facts, calculations, assumptions and recommendations. Give a source and calculation method for each figure. Mark unavailable information “unknown”.
Follow your organisation’s tool and data rules. Include only the permitted client, employee or commercial information needed for the task. Replacing names with codes does not necessarily make the material anonymous.
First pass: check comparability before interpreting results
Enquiries received this month may close next month. Dividing this month’s sales by this month’s enquiries may therefore not measure conversion of the same group.
Do not recommend business actions yet. Check whether the evidence can answer the decision question.
1. List each metric’s definition, unit, period, denominator and source.
2. Flag missing or duplicate records, definition changes, and groups or periods that cannot be compared directly.
3. Calculate only comparable measures, showing formulas and substituted values.
4. Identify up to three material gaps, explaining which person or record could resolve them.
Use sections for confirmed findings, non-comparable items and missing evidence.
Ask for auditable calculations and references. Independently check amounts, rates and totals against the source spreadsheet or another calculation.
Second pass: compare options, including doing nothing
A proposal containing only benefits gives a manager little basis for choosing. Supply at least two feasible options and the status quo.
Using the checked evidence, compare [option A], [option B] and the status quo.
For each, identify the supported potential effect, staffing/cost, dependencies, failure risks, reversibility and a way to test it.
Label unsupported benefits as hypotheses; do not estimate ROI without inputs. Exclude options that breach hard constraints before comparing the rest.
If scoring would help, propose criteria and weights for my approval. Do not substitute a total score for judgement.
Two options may each require 20 hours, but only one may be deferrable. Ask AI to show the displaced work and agreements needed, as well as the resource totals.
Third pass: identify evidence that would change your mind
Once you favour an option, test the case against it. Ask what would reverse the recommendation rather than requesting an open-ended list of risks.
I currently favour [option] because [reasons]. Identify:
the strongest objection;
known evidence inconsistent with the recommendation;
the unknown most likely to change the choice;
the smallest feasible check before starting;
and results that should trigger stopping or switching.
Label the evidence or verification needed for each point. Do not invent facts to create an objection.
Slower follow-up occurring alongside lower conversion is a reason to investigate, not proof of causation. A change in channels or customer mix may also matter. Treat AI-generated explanations as hypotheses to check.
Fourth pass: prepare a one-page decision brief
Only now ask for the polished version. Keep detailed analysis in supporting material and make the decision, evidence and trade-offs easy to find.
Use only the checked analysis to prepare this one-page brief:
Decision requested: [one sentence].
Evidence: [up to three findings with sources].
Recommendation: [why it is preferable to alternatives].
Trade-offs: [what will be delayed or not done].
Unknowns: [gaps material to approval].
Execution: [confirmed owners and dates; otherwise leave blank].
Review: [metric, timing and stop/adapt conditions].
Do not turn recommendations into approvals or add commitments.
Correction prompt: “You described slower follow-up as the cause of lower conversion. We only have changes over the same period. Retain the figures, label the explanation a hypothesis and list the case records needed to test it.” A precise correction is more useful than “go deeper”.
Describe the skill in an interview
If you are considering your next role, keep an example showing the decision, where AI helped, what you corrected and how you followed through. Include time savings or improvements only where you have records.
Answer structure: I needed to decide [question]. I prepared [data and definitions] and used AI to compare [options]. During review I found [issue], so I checked [evidence]. I recommended [action and trade-off] and followed it using [actual measure].
See the complete performance analysis and resource-prioritisation examples to put the prompts into practice.