Find one workflow worth changing
Think of one task you did this week that took more effort than it should. We’ll look at the steps together and find one small change you can try.
You’ll leave with a simple map of that task and a plan for testing one use of AI.
Before you start
Open your usual AI chat, such as ChatGPT or Claude. Choose a repeated task you know well: preparing a meeting summary, drafting a follow-up, or putting together a weekly update.
Have an example of the input and the finished result nearby. Replace the square-bracketed text in the prompts. For a first test, use made-up or anonymised material you’re comfortable sharing with the tool.
Describe what happens now
Give the AI the task as it happens today, including the awkward bits. You don’t need a polished process document.
Prompt 1 · Map the taskHelp me map a task I repeat. The task: [what I do] What starts it: [an email, meeting, request, or other input] My usual steps: [a rough list is fine] Who uses the result: [person or team] What a useful finished result looks like: [describe it] The part that takes the most effort: [where I get stuck] Ask one question at a time if you need more detail. Then map the steps in order. Mark where a person makes a decision or checks something. Do not recommend tools yet.
You should recognise your own task in the result. Add any missing handoffs, approvals, or exceptions before moving on. If the map describes an ideal process rather than your real one, correct it.
Choose one change to try
You don’t need to automate the whole task. Choose a step you can test and check without disrupting anyone else’s work.
Prompt 2 · Plan a small testUsing the task map we agreed on, suggest one step where AI could help. Explain what I would give the AI, what it would produce, and what I would still need to decide or check. Prefer a test I can run manually in this chat before connecting any tools. Write a reusable prompt for that step with clearly labelled placeholders. Suggest a small sample to test it on and two or three observable checks for a useful result. Name any information you are missing. Do not assume AI will save time.
Look for a test with a clear input, an output you can review, and a named human check. Try the suggested prompt on your sample. Keep the original beside the result so you can compare them.
Compare before you keep it
Review the result yourself, including the effort it took to correct it. Then use this prompt to think through what you learned.
Follow-up · Review your testHelp me decide whether this small AI test is worth repeating. What I tried: [the step] What worked: [specific observations] What I had to fix: [missing, wrong, or unhelpful parts] Time and effort compared with my usual approach: [what I observed, or “not measured”] Compare this with the checks we agreed on. Suggest whether to repeat the test, change one thing, or stop. Explain the tradeoff. Do not invent savings or assume the next test will work better.
You should have a practical next decision. If checking the output costs more effort than doing the step yourself, that is a useful finding too. Leave the rest of your process alone while you decide.
A small test, from start to finish
Illustrative example: turning meeting notes into a follow-up email.
- The task today
- After a meeting, you read your notes, pull out decisions, check who owns each action, and draft an email.
- The step to try
- Give AI a made-up set of notes and ask for a first email draft. Tell it to mark missing owners or dates rather than guess.
- What to check
- Every action comes from the notes. Names and dates match. Unclear items are flagged for you to resolve.
- What you decide
- If the draft is useful after a quick review, try a second sample. If it invents commitments or needs a full rewrite, change the prompt or keep doing that step yourself. You still approve and send the email.
If you get stuck
The AI suggests a whole new system
Ask it to return to one step you can test manually. You can explore connected tools later if the small test proves useful.
I can’t tell whether the result is better
Write down what a usable result needs to contain. Compare it with a recent example of your own work, including the time spent checking and fixing it.
Your final check
Tick these as you go. The ticks reset when you leave this page.
Keep going when you’re ready.
Pick another guide for the next problem you’re working on.