By Saber J.1 min read
What an AI project really costs (beyond the API)
Inference is the easy line. Data, access, run, and the person who stays until adoption usually dwarf it.
The unit price of an API call is easy to show. It is rarely the expensive part.
What actually eats the budget
Cleaning and rights on the corpus. Wiring into ticketing or the ERP. Hosting in a region you can defend. Logs. A person who stays until the tool is used. Then the monthly run.
What to write at kickoff
A use case you can measure. An inference ceiling. Who operates. What happens if usage doubles. If those lines are missing, the PoC was a demo with a nicer font.
We state a fee before we start. Most assessments stay under €2,000. The next mission is a separate decision.