AI That Works On Demand: Episode 2

Most people aren’t getting the results they expect from AI,  and it usually comes down to the prompt. 

In the second episode of our AI That Works webinar series, we broke down how to write prompts that work; so Copilot (and AI tools like it) become genuinely useful in day-to-day work. 

Watch the recording and you’ll learn: 

• A simple prompt structure anyone can use to get clearer, more accurate outputs 

• Real examples for everyday tasks (emails, summaries, tables and more) 

• How to build safe use into the process, so your team can adopt AI with confidence 

Key Takeaways:

AI That Works – Episode Two – Prompting That Gets Results

Why AI feels inconsistent (and how to fix it)

When people say “AI is rubbish”, what they often mean is: “I asked a vague question and got a vague answer.”

AI works best when you treat it like a new starter in your business. If you don’t give it context, it can’t read your mind and the output will be generic.

The solution isn’t “better AI”. It’s better prompting.

The CRAFT framework: a simple structure for prompts that work

To make prompting easy and repeatable, Fraser Dear, our head of AI and data innovation introduced CRAFT, a framework you can use in any AI tool:

1) Context

What’s happening? Who is this for? What matters in this situation?

2) Role

Who should the AI act as? (e.g., a comms advisor, HR lead, analyst, project manager)

3) Action

What do you want it to do? (write, summarise, create options, extract actions, draft a table)

4) Format

How should it be structured? (bullet points, short paragraphs, table, headings, word count)

5) Tone

How should it sound? (confident, friendly, professional, concise, positive)

This one shift, adding structure to your prompt, is what turns generic outputs into something useful you can actually work with.

Real examples: prompts for everyday work

The session included live demos showing how the same approach can be used for common business tasks, including:

• Drafting a short email with a clear purpose and tone
• Summarising a long document into exec-friendly bullet points
• Creating a simple table of risks, mitigations and recommendations
• Turning source content into a briefing-style summary with clear headings

Copilot, not pilot: iterate to get to the right result

A big theme in Episode 2 is mindset.

AI isn’t a “one and done” tool, it’s a co-worker.

Start with a solid prompt, read the output, then refine:

• Add more detail
• Tighten the format
• Change the tone
• Ask it to rewrite for a different audience

Small changes to the prompt often create big changes in the output.

Safe use is part of good prompting

The session follows on from Episode 1 of AI that Works and reinforces the importance of using AI tools with the right protections in place (for example, Enterprise Data Protection in Copilot Chat), and being intentional about what you share, where it goes, and who has access.

In short: safe use isn’t a separate “AI policy” document. It’s a habit you build into everyday work.

Quick wins you can try this week

If you want to improve results immediately, try these:

• Use CRAFT once today for a real task you’re already doing
• Save your best prompts so you can reuse them (and share them internally)
• Use a prompt gallery as a starting point, then tailor it to your role and context
• Spend 30 seconds improving the prompt, it can save minutes (or hours) of editing later

What’s next

Episode 3 – Small AI wins: Everyday Hints and Tips Hidden in Your Work Tools

Register here.

Next Steps

If this session sparked ideas about how AI could reduce busywork in your organisation, we’d love to help you take the next step.

Explore our wider AI services, from secure Microsoft Copilot adoption and adoption support, to automation, custom AI agents and data-readiness programmes.