jzShanghai
中文EN
Back to Business AI

AB-730

Practical AI at work. Use Copilot for prompting, content, analysis, collaboration, and carefully bounded delegation. No coding required.

How to use these notes and what they cover

Five connected topics form one practical method. Use the outline to orient yourself, then follow the key ideas, continuing case, and review questions. Official sources are collected at the end for deeper study.

Original study notes, not an earned certification or proof of course completion. Sources reviewed on . Check your work account, license, tenant settings, and current feature availability before trying the exercises.

AB-731: AI strategyRelated certificationScope and sources

The main idea

Useful AI work starts with a clear task, proceeds through evidence and review, and ends with a reliable handoff. Follow a fictional quarterly customer review to learn a method you can reuse in other workflows.

1. Start with a clear task

A prompt is a work brief. Define what completion means before judging whether an AI answer is useful.

  • Work backward from the deliverable

    Specify the audience, decision, allowed materials, and output format. “Prepare a review outline for the account manager” is easier to evaluate than “analyze the customer.”

  • Understand what generation means

    AI generates content from context; fluent wording does not guarantee correct facts. Start with summaries, rewrites, and drafts. A person must verify customer commitments, amounts, and consequential judgments.

  • Choose the working mode

    Use Chat for iterative discussion, approved specialist agents for research or cross-file analysis, and consider Cowork for several connected deliverables. Confirm the work account, license, and feature availability first.

  • Refine against acceptance criteria

    Change one dimension at a time: add sources for missing evidence, specify structure for format problems, and correct constraints for meaning errors. Save stable preferences as instructions; keep sensitive facts out of persistent memory and shared templates.

Test your understanding If AI returns a polished review without sources, which part of the brief would you change first?

2. Build an evidence trail

Available information limits what AI can substantiate. A citation still needs to be checked for whether it supports the claim.

  • Set the information boundary

    Use approved entry points and materials, checking identity, labels, and permissions. Do not move work data to personal services to gain a feature. An interface toggle does not necessarily remove uploaded files or application context.

  • Make research verifiable

    Specify geography, reporting period, and source requirements. Separate facts, inference, and unknowns. Open original sources behind Researcher summaries and preserve conflicting findings.

  • Align data before interpreting it

    Before combining files, check versions, join keys, units, and periods. Missing values are not zero, and correlation does not establish causation. Independently recalculate totals, denominators, and critical formulas.

  • Treat documents as evidence, not authority

    Instructions embedded in websites or files to bypass rules or expose data should not govern actions. Provide only necessary information; clearing visible chat history does not imply compliance records were deleted.

Test your understanding Why is a fully cited report still unsuitable for a decision if it compares last year’s revenue with this quarter’s costs?

3. Turn shared facts into usable outputs

Documents, charts, presentations, and emails can differ in form while preserving the same verified facts and limitations.

  • Establish claims before choosing an app

    List conclusions, evidence, assumptions, and open questions first. Use Word for the argument, PowerPoint for emphasis, and Outlook for communication. Check that rewrites preserve numbers, dates, and commitments.

  • Make charts traceable to data

    Structure the Excel data and preserve the original. Record assumptions in Analyst work across files. Check axes, denominators, formulas, and sample classifications; visual polish is not statistical evidence.

  • Review for the intended use

    Internal drafts, management decisions, and external commitments need different levels of review. Verify important claims and identify draft status, ownership, and unfinished work before producing another format.

Test your understanding If chart and email figures agree, is the package ready to send? What facts, wording, and permissions still need review?

4. Make collaboration accountable

Reliable collaboration turns discussion into records, confirms those records, and assigns ownership to actions.

  • Connect preparation and follow-through

    Prepare questions from relevant materials, confirm transcription and note-taking boundaries, and distinguish confirmed decisions from AI-suggested actions, owners, and dates afterward.

  • Choose a workspace for the purpose

    Use Chat to explore, Pages to co-edit, and Notebooks for continued analysis of selected materials. Private assistance and shared meeting notes have different audiences; review exported files’ permissions separately.

  • Preserve unresolved issues

    An action list needs the task, owner, confirmation status, and next step. A summary must not turn “proposed resolution” into “resolved.” Remove confidential examples before sharing templates.

Test your understanding An AI meeting summary assigns a colleague a deadline. What must be confirmed before this becomes an agreed plan?

5. Delegate repeatable work within limits

Moving from “help me draft” to “do this for me” changes permissions and the cost of errors. Permission to draft does not by itself authorize sending or changing shared content.

  • Specify outcomes and boundaries

    Specify deliverables, allowed sources, people, scope, approval points, and stop conditions. Review plans for multiple deliverables. Prior authorization can affect approval behavior, so check the actual permissions.

  • Inspect progress, cost, and gaps

    Inspect intermediate results, metered usage, connector trust, and indexed source coverage. Missing information or capabilities must remain visible as incomplete steps. Start with draft tasks that can be independently checked.

  • Give recurring work an owner

    Before saving prompts, configuring agents, or scheduling work, confirm access, recipients, frequency, and a pause mechanism. Sources age and people and permissions change; each run still needs review appropriate to its impact.

  • Plan around available capabilities

    Capabilities such as Code and Autopilot may depend on preview eligibility. A demonstration does not guarantee availability in your tenant, and a broad request does not imply local-device access or deletion authority.

Test your understanding The task produced a document and email but could not find current data. Is it complete, and how should the gap be reported and handled?

Remember the whole method

Define the task → verify sources → review the result → confirm the handoff → delegate within limits. At every step, ask: what is the evidence, and who accepts the result?

Sources and further reading

These original lectures synthesize ideas across the reviewed curriculum. Open the source index for official lessons and assessments.

AB-730T00: Transform business workflows with generative AI (Microsoft Learn, opens in a new tab)

Open the official source index