Pick the right tasks for AI

Not every task should go to AI, and the fastest way to waste a month is to start with the wrong ones. A task is a good AI candidate when it is language-heavy, repeats often, has a forgiving first draft, and you can judge the quality of the result in under a minute. Writing follow-up emails, turning messy notes into a checklist, summarizing a long thread, rewriting the same offer for three channels, and drafting job descriptions all clear that bar easily. A task is a poor AI candidate when it depends on facts only you hold, when being wrong is expensive, or when the value comes from human presence. Quoting a job without seeing the site, deciding whether to fire someone, calculating payroll tax, or sending an unreviewed message to your largest client are not AI tasks. They can still use AI as an input — a summary of the file, a list of questions to ask — but the decision and the send stay with a person. The practical filter is a two-by-two: how often does this happen, and how bad is a mediocre first draft? High frequency plus low blast radius is where you start, because you get many reps quickly and mistakes are cheap. Low frequency plus high blast radius is where you stop. Most small businesses have between five and fifteen tasks in the top-left box, and working through them in order of frequency produces more time savings in a month than any single clever prompt. Owners often start with their hardest task because that is what hurts most, and then conclude AI does not work. Hard tasks usually depend on private judgment and context, which is exactly what the tool lacks. Starting with the boring frequent tasks feels less impressive and produces far more recovered time. A useful tiebreaker is whether you could judge the result in under a minute. If checking the output takes longer than doing the work, the task is a poor candidate no matter how repetitive it is. A dental office listed twelve weekly tasks and found that three of them — reminder texts, post-visit care notes, and insurance clarification replies — accounted for six hours a week. Those three went first, and the practice never needed to touch the harder scheduling decisions to get most of the benefit.

Key takeaways

• A task is a poor AI candidate when it depends on facts only you hold, when being wrong is expensive, or when the value comes from human presence.
• The practical filter is a two-by-two: how often does this happen, and how bad is a mediocre first draft?
• Reminder rewriting is frequent, language-heavy, and cheap to correct.

Sign in to complete activities, save progress, and use the AI Coach.