Course 3 - Module 5 - Lesson 2
Capstone: design a practical AI automation plan with first, next, and later steps
Your AI Opportunity Map
Write a plan someone could actually run
Your capstone is a practical automation plan for one repeated task. State the problem, owner, current time and quality, available input, desired output, and constraints. Pick a task you can trial with fictional or approved material during the next week.
For example, a club coordinator spends thirty minutes making a weekly equipment summary. The target is a checked draft in fifteen minutes, with every quantity traceable to the supplied inventory. The coordinator reviews the result, and the routine cannot place orders or contact members.
Separate first, next, and later
First, write a reusable brief and test three sample inventories: complete, incomplete, and contradictory. Define the expected summary for each before running the trial. Keep the source beside the draft and record preparation time, corrections, and any invented quantities.
Next, if the drafts are consistently useful, connect a read-only export from the existing inventory tool and produce a preview. Later, consider scheduling the draft or adding a review queue. Each expansion needs an owner, an access decision, and a way to stop or recover.
Make the decision depend on results
Choose success criteria before the pilot: perhaps ten checked runs, no invented quantities, and at least fifteen minutes saved per run after review. Include setup and ongoing costs when judging value. Decide how many failures will trigger a pause and who investigates them.
Present a one-page plan containing the workflow, sample tests, measurements, boundaries, and fallback to the existing manual method. End with a dated review point and a decision: continue, revise, or stop. A realistic plan makes the next action clear without promising benefits you have not measured.
Tiny experiment
Draft your one-page capstone with first, next, and later steps. Include three sample tests, one measurable quality target, a review date, and a manual fallback.
Confidence check
What pilot result would make you stop or redesign the routine even if it appeared to save time?