Where AI saves a small business money (and where it quietly costs you)

• Judgement work dressed up as repetitive work. Pricing an unusual job, handling an upset customer, deciding who to hire. If the right answer depends on context only you hold, automating it produces confident nonsense.
• One-off tasks. If something happens twice a year, the setup cost will never be repaid. Do it by hand and move on.
• Platform sprawl. Buying an all-in-one suite to solve a single problem is how businesses end up paying for four tools that overlap. You very often already own something that does the job.

Take any task that annoys you and answer four questions about it:

• How many hours a week does it take, across everyone who touches it?
• Does it follow the same steps almost every time?
• If it went wrong, would somebody notice before a customer did?
• What would you do with the time if it disappeared?

A task with real hours, repeatable steps, a safety net and a better use for the freed time is worth fixing. If the last answer is "nothing in particular," the saving is theoretical and the project will disappoint you. That question weeds out more bad projects than any technical review.

Ten hours a week at a $35 loaded hourly cost is roughly $18,000 a year of capacity. That is the number you compare a tool or a project against — not the monthly subscription in isolation. A $200 month that returns six hours a week is obvious. A $50 month that returns twenty minutes is not worth the account you have to manage.

Write the before number down. Response time, hours, dollars, missed inquiries — whichever fits. Every disappointing automation project we have been asked to rescue had one thing in common: nobody recorded what things looked like beforehand, so nobody could tell whether it worked.