It is the first question in almost every conversation, and it is the wrong shape. "How much does AI cost" invites a number, and a number without context is how people end up paying for software nobody opens.
The software is rarely the expensive part
Three things carry the cost, and they are not equal.
- Working out what to automate. This is most of it. The task has to be described precisely enough to hand over, which usually means writing down a process that currently lives in somebody's head.
- Running it. Modern models are billed per use, in fractions of a cent per request. For a business your size this is normally the smallest line on the page — and it scales with how much you use it, which is the right direction for the risk to run.
- Owning it afterward. Anything nobody on your team understands has a maintenance cost that shows up later, usually at the worst time.
The one number to know first
How many hours a week does the task take today, and who does them?
Without that, nothing else can be judged. A tool that costs money is not expensive or cheap in the abstract — it is expensive or cheap against the hours it removes. Most of the automation we turn down gets turned down because that number came back smaller than the owner expected, and we would rather say so early.
Why we quote fixed
You get a fixed proposal before any work starts, because "it depends" is not something you can budget against. If the assessment says the honest answer is not yet, that is a legitimate result and it costs you nothing.