Before you buy another AI tool, ask these five questions.

Start with last Tuesday, not a wishlist.

Think about an actual task you completed recently. What triggered it? Where did the information come from? What did you have to decide? Write down the steps before naming a tool. “I need help with productivity” is too broad to evaluate. “I rewrite the same onboarding instructions for each new client” is a workable starting point.

1. What job is this replacing?

Define one job in a sentence. For example: “Help me turn my approved project notes into a draft status update.” A tool that cannot be connected to a specific job belongs on a research list, not automatically in your budget. Ask whether the job should exist at all before automating it.

2. Can the tools I already use do it?

Check your existing assistant, document system, calendar, and task tracker. Try a simpler template or native reminder. The decision is not whether the new tool has more features. It is whether those features address a real limitation in your current process.

3. What still needs a person?

Mark decisions, approvals, and external commitments. Drafting a reply is different from sending one. Proposing a schedule is different from moving a client meeting. Write down where you must stay involved and make that review visible in the workflow.

4. What will ownership cost?

Include setup, subscriptions, usage, review, troubleshooting, and training. A workflow with a low subscription price can still consume time. Estimate these costs as assumptions, then replace them with observations from a small test.

5. How will I know to keep it?

Choose a short trial with a baseline. Track manual effort, review effort, missed steps, and whether the output was usable. Decide in advance what would justify keeping, revising, or removing the tool. Time saved is capacity, not automatically cash saved.

Try this today

Pick one subscription or proposed purchase. Write its primary job, your existing alternative, its review requirement, and your test. Give the decision a date. You are building a useful system, not trying to win a software-collecting competition.