Case study · AI literacy
A large Canadian pension and investment organization rolled out enterprise AI tools to its whole workforce and needed a short course that would still be accurate a year later. I took it from client consultation through storyboard, Rise build and two motion graphics videos.
Client name and internal tool names are withheld. The learner scenarios and design decisions are shown as delivered, and client names are removed from screenshots.
The client had rolled out Microsoft Copilot and an in-house AI assistant, but had no foundation course: one short module every employee could take to understand AI, prompt it well and follow the organization's rules. The source was a 35-slide internal deck. The brief asked for 20 minutes, three objectives, knowledge checks throughout, and content that would stay useful as the tools changed.
AI moves faster than e-learning. The real challenge was a course about AI that wouldn't be out of date on launch day.
I built this activity with Rise's Scenario block, using its character, dialogue and choice structure to put learners in Reese's seat. Instead of recalling a definition, they make the decisions Reese faces at work, one GCSE element at a time: Goal, Context, Source and Expectations. Each choice branches to its own feedback. Stronger choices move the scenario forward, and weaker ones explain what's missing and send learners back to try again.
Here it is rebuilt for the web. The prompt assembles as you go.
Reese needs to send a weekly update email to the team and wants Copilot to help draft it. Help Reese build the strongest prompt.
Goal Context Source Expectations
Reese sends it, and the draft comes back focused and professional, ready to use after a few small edits.
The same activity in the finished course, built with Rise's Scenario block:
Durable ideas carry the course: what generative AI is, four responsible AI principles, the GCSE framework and review-before-you-use. Tool features sit in one dated section, and the course ends by pointing to the live learning hub where those details stay current. Before storyboarding, I also asked how this course differed from an existing governance course. Nobody had flagged the overlap, so the client added their governance team as a reviewer.
Every scenario is a moment in the day of Reese, a project coordinator who has tried AI but isn't sure what's allowed or whether to trust it. The course ends at the same desk with the same to-do list, now handled with confidence.
Every distractor is something a reasonable employee might actually do, and the feedback explains why it falls short.
| The tempting answer | Why it's wrong |
|---|---|
| "It's an approved tool, so I can paste in the whole meeting transcript." | An approved tool doesn't make every input appropriate. |
| "I'll ask the AI to check its own summary." | AI often confirms its own mistakes. A person does the final review. |
| "External meetings should never be transcribed." | An overcorrection. Ask participants first and respect any objection. |
I scripted, storyboarded and produced two short motion graphics videos. Both use a generic chat window instead of a real product's screens, so a tool redesign won't date them.
Imagine you're prepping for an important meeting. You open your AI assistant and type: "Help me with my meeting." The response is generic: tips for running meetings, advice on agendas. None of it is what you actually needed.
So you try again: "Summarize the key updates so I can brief my team at our meeting tomorrow. Use the project status report I uploaded. Focus on milestones, any delays, and next steps. Keep it concise and use plain language." The response is completely different. Focused, relevant, and actually useful.
Same tool. Same task. Completely different result. How you communicate with AI determines what you get back. Vague input, vague output. Clear, specific input gives you something you can actually use.
That's prompt engineering: crafting your requests thoughtfully so the AI gives you the most accurate, relevant, and useful response.
A strong Copilot prompt comes down to four parts: Goal, Context, Source, and Expectations. GCSE. Together, they form a simple framework that gets you better results from any AI tool.
Your Goal is what you want the AI to do. Be specific about the task and the outcome.
Your Context is the background that helps the AI understand why you need it and who's involved.
Your Source is what the AI should draw from. Point it to the right document, data, or reference so it works from the right foundation.
Your Expectations define how the AI should respond: the format, tone, and length.
Goal. Context. Source. Expectations. Four parts. One clear prompt. And a result that works.
The work that mattered most was human: the client questions, the durable-versus-dated split, Reese, and choosing which mistakes to teach. AI sped up the rest. Only material the client was comfortable sharing went into an AI tool, and nothing confidential did.
Turned the 35-slide deck into an outline mapped to the objectives, using Claude, ChatGPT and Gemini (the client's own tool).
Checked it against the policy slides and cut what the governance course already covered.
Generic examples that sounded plausible but didn't match real work.
Asked the SME for real situations. They became Reese's scenarios.
Midjourney images, ElevenLabs voiceover, Canva animation.
Wrote the scripts and motion storyboards, directed pacing and checked every image for brand fit.
Built to WCAG 2.2 AA and AODA from the storyboard on: transcripts for every video, alt text planned per image, descriptive link text, brand colours checked for contrast, and keyboard and NVDA testing in Edge and Chrome.
Summarized from learner feedback, not direct quotes.