How to choose and test your first AI workflow
Start with a task you repeat often, where the input is easy to describe and you can quickly tell whether the output is right. Try it on one redacted or fictional sample. Check the result against the source, then test a fresh sample before deciding whether to use it again.
Try it on your own task
- Name one input and one output. For example, turn each set of meeting notes into a list of decisions and open actions.
- Write down the rules you use today. Tell the tool to use only the supplied notes and mark a missing owner or due date as “to confirm”.
- Compare the draft line by line with the notes. Correct anything missing or invented, then try the same instructions on a fresh sample.
- Compare your usual time with the time spent preparing, drafting, checking and correcting the AI version. Count only drafts that meet the same standard.
Worked example
Fictional scenario
- Source
- A studio’s notes say that Mina will ask for a supplier quote. No deadline was agreed. The team also discussed a Friday launch but made no decision.
- Before review
- Mina will obtain the quote by Friday. Launch on Friday.
- After review
- Action: Mina to request the supplier quote. Due date: to confirm. Launch timing: discussed, decision pending.
Check your result
- Does every action or decision trace back to the notes?
- Are unknown owners and dates flagged for a person to resolve?
- Did the second sample still meet your quality standard, and how much time did the complete review take?