Start with work, not a tool
For one week, note the work you repeat: sorting incoming inquiries, drafting quotes, summarizing meetings, or moving information between systems. Record how often it happens, how long it takes, and what goes wrong. Do not put customer or employee information into a new AI tool while experimenting.
Choose a small, reversible task
A good first task happens often, has a clear input and output, and can be checked by a person. Drafting a follow-up for a person to review is usually easier to evaluate than giving a system permission to send every reply. Avoid starting with hiring decisions, legal advice, payments, or other decisions with serious consequences.
Write your one-page brief
Complete these sentences: “When ___ happens, our team currently ___.” “A useful improvement would ___.” “A person must approve ___.” “We will measure ___.” Add who owns the process, what information is required, and how to return to the existing process if the pilot fails.
Measure before and after
Capture a baseline over a representative week. Compare time spent, corrections required, and completed work during the pilot. Include subscription costs, setup time, human review, and maintenance. A faster draft that needs extensive correction may not save time.
Decide whether to expand
Keep the workflow only if the output is reliable enough for the task and the complete cost makes sense. Document when it needs human review and who can stop it. Expand to a second workflow after the first one is useful in everyday work.