The costs that are not on the proposal

Most AI proposals follow a familiar shape. The largest share of the money goes to licences, a smaller share to connecting things up, and a sliver to helping people adopt it. Nearly all of them underestimate what it takes to make AI work in a real organisation. Seven costs rarely appear on the proposal and often appear later.

Usage at scale. Trying something is cheap; using it every day is not always cheap. Many tools charge by how much they are used, so a feature that cost very little in a trial can cost many times more once the whole team relies on it. Work out the cost per customer, per document or per person before the rollout, not after the first surprising bill.

Getting the information ready. Before AI can do anything useful, someone has to tidy, organise and check what it will read. This work is invisible in a demo and very real afterwards. It can cost as much as the AI itself. Then connecting things: “just connect it to our customer database” hides weeks of access, permissions, failure cases and odd exceptions. Projects stall here more often than anywhere else.

People. The people who will use it decide whether it works. If the team does not trust what the tool produces, they will not use it, and then it does not matter how good it is. Give this a real share of the budget and real time. Then looking after it: a new kind of request turns up, instructions that worked last month produce something odd this month, the maker updates the tool and it behaves differently. AI in daily use needs someone who owns it.

Rules and paperwork, which are not getting lighter. And depending on one vendor: introductory prices rise in the second year, and many AI vendors are young companies. Keep your own information in your own systems, and avoid building your work so tightly around one tool that you cannot leave.

Add these up and the first-year cost of an AI project is often several times the licence price. That is not a reason not to begin. It is a reason to plan honestly, and to start small enough that the surprises stay small too.