The AI budget: what a thirty-person company should actually be spending
Vendors will not tell you, because the honest number is small. Here are the three lines and roughly what each should cost.
Nobody publishes realistic AI budgets for small companies, because the realistic figure is unimpressive and nobody sells anything by quoting it. So here is the honest structure, in three lines.
Line one: seats for people
Ordinary business accounts for the people who will actually use a chat tool daily. In most SMEs that is not everyone — it is the office, sales, and whoever writes for a living. Ten to twenty per cent of headcount is a normal starting point, expanding as use proves itself.
Add one or two premium seats for the people doing contract comparison or complex planning, as covered in December. Not everyone. One or two.
Line two: usage inside systems
This is the line nobody forecasts and the one that surprises. When AI runs inside a process — reading invoices, classifying mail — you pay per document rather than per person.
It is usually smaller than people fear and harder to predict than people like. Estimate it as: volume per month, times the tier you assigned in April, and then double it for the first three months while the configuration settles.
Line three: the one that is actually large
Internal time. Someone cleaning master data, someone defining the process, someone checking output for the first two months.
In every SME project we would call successful, this line is several times the software cost. It does not appear in any vendor proposal, and pretending it is zero is the single most common reason a project overruns.
The sanity check
If your total software spend on AI exceeds what you pay for your accounting package, stop and ask why. There are good answers — high-volume document processing is one — but there should be an answer.
And if a proposal puts the software cost in the tens of thousands for a company of thirty, the thing being sold is a project, not a tool. That may be what you need. It should be a conscious decision rather than a surprise on the second invoice.
What this means in practice
Start smaller than feels serious. A handful of seats, one automated process, and an honest allocation of internal time. Expand what demonstrably works and cancel what does not, which is possible precisely because you kept the contracts short.