Ask me which AI tool is best for a small business and the honest answer is: the one you are still using in six months. Not the most capable one on the list. Usually it is the simplest thing that fixed a real annoyance.
That sounds like a throwaway line, but it is the only test that matters, and it is a harder one to pass than it looks. Plenty of automations work perfectly on the day they are built and are quietly switched off by the spring. Nobody announces it. The thing just stops being part of how the week runs, and the subscription carries on coming out of the account.
Three things decide it, and none of them are about the technology.
It has to replace something that was actually costing you
The first question is not which tool. It is whether the job is worth removing at all.
Work out what the task costs you now — hours a week, times what an hour of that person's time is worth — and compare. If a task takes two hours a week and the build is $495, the arithmetic is quick. If it takes ten minutes a week, do not automate it. Ten minutes a week is not a problem, it is just something you do, and paying to remove it is how people end up with software they resent.
The same logic applies to what you are already paying for. There is a calculator on this site that adds up your monthly software and outside services, and it has a section built to shrink the total on purpose — accounting, bookkeeping, payroll, your industry software. You tap those so they never show up as something to cut, because they are not the problem. What is usually left over is more interesting: five separate bills and five logins doing one job, with nobody owning the join between them.
Most tools of that kind are built to make the number look as big as possible. A number built to be honest is more useful, even when it tells you to leave things alone.
A person still sees anything the customer sees
The question I get asked most often, usually about halfway through a first conversation, is some version of: is a robot going to be talking to my customers?
It is a fair question and the answer should be no.
AI is confidently wrong sometimes, and it does not know it. So nothing goes to a customer without a person seeing it first, and nothing touches your money or your records without a rule around it. That is not a reassurance about how good the tools have got. It is a scoping decision, made when the work is specified, and it is not a switch you flip afterwards.
It is also, in practice, the difference between an automation you are still using in six months and one you quietly turn off after it embarrasses you once. A single wrong message sent to a customer in your name costs more trust than the automation was ever going to save you in time. Anyone telling you to just let it run has not had to clean up after it.
It should be your own decisions, just on a timer
Here is one of the things I actually build, described plainly, because people picture something stranger than it is.
A quote that follows itself up on day three. No AI writes to your customer. It is the follow-up you already wrote, sent on day three if nobody has answered — the one that currently waits on somebody remembering. You approve the wording once. After that it goes out on schedule, whether or not the week got away from you.
That is the shape of the work that lasts. It is not a system making judgement calls on your behalf. It is your own judgement, captured once, and then repeated reliably. Nobody has to trust anything new for it to keep working, which is exactly why it is still running in six months.
What that means for buying one
One automation is $495 flat and takes about a week. Two or three that hand off to each other is $1,500. The price is agreed before any work starts, so there is no invoice you have not already agreed to, and you own what gets built, including the accounts it runs on. I do not resell anybody's software or take a cut of it, and everything comes with a one-page description of what it touches and how to switch it off.
That last document is the one people skip and it is the one that matters here. If your team cannot explain what a thing does, they will not trust it, and it will quietly stop being used.
And if the arithmetic does not work, I will say so on the call. That is not modesty. An automation nobody wanted is worse for me than a job I turned down, because it is the one you remember.