Owners of small and mid-size service firms are not short of advice about AI. They are short of a starting point that does not require a six-month project or a new department. The useful starting point is almost always a single process the owner still does themselves, late in the day, because nobody else will.
This is a best-practice note from Two Dogs Advisory, a fictional firm. It is written for illustration, not as advice to a real client.
Look for work that already has a script, even if the script is only in the owner’s head. Intake questions, appointment confirmations, quote follow-ups, invoice reminders — these are good first candidates because the “right” answer is usually already known. Strategy, hiring, and the unusual job are not. An agent is a poor choice for work the owner could not explain to a new employee in fifteen minutes.
Write the process down before you buy a tool. Who starts it, what information has to be true, what happens in the usual case, and what happens when it is not usual. If you cannot write that down, the agent cannot either. The first week of work is almost never a software demo. It is a conversation about how the firm already runs.
Then run a narrow test beside the current way of working, not instead of it. One form, one service line, one follow-up sequence. The owner should still see every message or booking at the start. The question to answer is not “did we adopt AI.” It is “did this take work off my plate without making a mess I then had to undo.”
A dental clinic might start with new-patient intake. An HVAC contractor might start with reminder texts before a visit. A small accounting firm might start with the scheduled requests for documents before tax season. The trades are different; the test is the same. Pick the process, write the rules, watch it for a few weeks, and only then decide whether it is worth extending.