← Back to Microsoft 365 Prototypes

Case study · AI adoption

Copilot in Excel: Prompt It, Then Prove It

AI skills training that builds a verification habit.

An organization wanted staff using Copilot in Excel. The real risk wasn't getting an answer; it was trusting a wrong one. I designed a 6-lesson course where every skill ends with a check, then prototyped it in two Articulate formats so the client could compare before development.

6
Lessons, the same in both formats
2
Articulate formats prototyped for the client to compare
~50 · ~35
Minutes in the Rise and Storyline versions
80%
Passing score on the Rise final assessment

The ask vs. the real goal

The ask

Teach staff to use Copilot in Excel for formulas, data cleanup, analysis, and charts.

The real goal

Build a verification habit: prompt it, then prove it, every time, before a result is shared.

People centered

Safe practice, no risk.

  • One fictional budget workbook for every example
  • No license, no real data, no broken files
  • Built for staff who build budgets and reports
Strategy driven

Judgment, not just prompts.

  • One method repeated in every lesson
  • Choices between options, not just clicks
  • Match how much Copilot can change to the risk
Impact focused

Fewer confident wrong answers.

  • A “Prove it” check after every practice
  • Five habits that keep results accurate and data secure
  • A final assessment in the Rise version

From the risk to two prototypes

In a budget or a shared report, a confident wrong answer costs more than the time Copilot saves.

  1. Hear the ask Teach Copilot in Excel for formulas, data cleanup, analysis, and charts.
  2. Name the real risk Trusting a wrong answer, so the course had to build a verification habit.
  3. Design one method Describe it, Copilot writes it, you prove it, in every lesson.
  4. Map six lessons Formulas, organizing data, analysis, charts, multi-step builds, and best practices.
  5. Prototype in two formats Built the same course as a Rise prototype and a Storyline prototype.
  6. Let the client compare The client clicked through both and chose the experience that fit their learners, before development.
  7. Test AI authoring Rebuilt the course in Mindsmith from my outline, as a labeled experiment.

Step 2 is the turning point.

Prompt it, then prove it, every time.

The habit behind the course

The client chose the format before development

Articulate Rise prototype
  • Scrolling lessons with a course menu
  • Select-a-prompt practice in a sample workbook
  • Maria's scenarios and a find-the-mistake check
  • Wrap-up with a 10-question final (80% to pass)
  • About 50 minutes
Open the Rise prototype →
Articulate Storyline prototype
  • Slide-based with a seek bar and menu
  • Type or select prompts in a simulated Copilot pane
  • Drag-and-drop sorting and a hotspot tour
  • Knowledge check at the end of every lesson
  • About 35 minutes
Open the Storyline prototype →

A third take: AI authoring experiment

I also rebuilt the course in Mindsmith, an AI course-authoring tool, starting from my own outline, to test how far an AI authoring tool can take a well-designed course. The result is an 8-section lesson with a timed “one-minute decision” scenario, tabs comparing Copilot's chat, plan, and edit modes, and a closing activity where learners write the prompt and the verify check they'll use this week.

Try the Mindsmith version The same course, generated with AI from my outline and refined in Mindsmith.

Six lessons, each with practice and a check

The same six lessons in both formats. The Rise version adds a wrap-up and final assessment.

Lesson 1

Prompt for Formulas

Tour the workspace, write and explain formulas, and troubleshoot errors.

Lesson 2

Organize Your Data

Highlight, sort, filter, format, and clean data with prompts.

Lesson 3

Analyze Your Data

Ask questions of the data, then refine the answer with follow-up questions.

Lesson 4

Visualize Your Results

Choose the right chart, then build charts for reports.

Lesson 5

Build Multi-Step Workbooks

Choose between chat, plan, and edit modes, then practice a multi-step build.

Lesson 6

Copilot in Excel Best Practices

Five habits that keep results accurate and data secure, then key takeaways.

Who did what

I designed the course and the judgment it teaches. AI tools built and generated what I directed.

Jacqueline Davis

Learning strategy & design

  • Named the real risk and the verification habit
  • Designed the lessons, practice activities, and assessment
  • Wrote the course outline used for every version
  • Reviewed every lesson

Client

Organization adopting Copilot

  • Set the goal of staff using Copilot in Excel
  • Compared the Rise and Storyline prototypes before development

AI tools

At my direction

  • Claude & Claude Code: built the Rise-style and Storyline-style prototypes
  • Mindsmith: generated a version of the course from my outline

Seven decisions and why

Show all 7 decisions
#DecisionWhy
1Teach verification, not just promptingA confident wrong answer costs more than the time Copilot saves
2Repeat one method in every lessonRepeating one habit in different tasks makes it automatic
3Practice in a simulated workbookSafe practice with no license, no real data, and no risk of breaking a file
4Make learners choose, not just clickChoosing between options builds judgment a prompt list can't
5Teach how much Copilot should changeMatching control to risk is the judgment the job requires
6Prototype in both Rise and StorylineThe client chose the format that fit their learners before development
7Test an AI authoring tool, and label itSee how far AI authoring can take a well-designed course, openly

Directed by a person, built with AI

Human authorship

The goal, the method, lesson design, practice activities, assessment, and review.

AI generation

Two prototypes built by Claude Code for a fictional company, Acme Financial, and one AI-authored version in Mindsmith.

Open by design

A course about checking AI's work should show its own. How I use AI

Try the Storyline prototype Type prompts in the simulated Copilot pane, answer each Prove it question, and drag tasks to the right mode.

Rolling out AI tools to your team?

Let's talk about how your people can use them well, and know when to double-check.

Let's talk