| Department | Starting task | Required input | Human check |
|---|---|---|---|
| Marketing | Turn a brief into content options | Approved brief and brand guidance | Check claims, audience and tone |
| Sales | Draft a meeting follow-up | Approved notes and agreed next steps | Verify every promise before sending |
| HR | Explain an internal policy | Current policy and intended audience | Verify rights and escalate individual cases |
| Operations | Document a recurring process | Verified steps and actual exceptions | Check sequence and exception handling |
| Finance | Draft a management commentary | Validated figures and period definitions | Recalculate figures and challenge explanations |
| Customer service | Draft a reply from a knowledge base | Approved answer sources | Check policy and escalation needs |
| Product | Organise interview observations | Authorised notes with context | Separate observation from interpretation |
| Legal | Compare versions of a clause | Authorised documents and jurisdiction | Expert review of meaning and omissions |
How should you choose a first task?
Prefer a bounded task that someone already understands. A short draft or internal summary is easier to evaluate than an autonomous process spanning several systems. Choose a representative example and write down what the reviewer must check.
What makes a task unsuitable for an initial workshop?
Pause when the source is unreliable, no one can evaluate the result or the output immediately affects a person, payment or external commitment. Resolve those conditions before introducing automation. A less ambitious first task often teaches more about the organisation’s real constraints.
How can you compare two candidate use cases?
Compare the frequency of the task, available evidence, sensitivity of the data and cost of checking the output. Avoid inventing an impact score without observation. After a pilot, compare the total effort including preparation, review and correction, not generation time alone.
