Most AI agencies start building on Monday. It’s understandable: building is visible, demonstrable, sellable. The problem is that systems projects — I’ve spent twenty years watching this with ERPs — don’t fail in construction. They fail at the framing.
What happens when you build without a plan
A typical project with no prior diagnosis: a tool is chosen before the process is understood, documentation is indexed that should never leave certain permissions, something that needed human judgement gets automated, and three months later the system technically works but nobody uses it. The cost is not just the project: it’s the distrust left behind towards any subsequent project.
What should happen first
Before building anything, there are questions that can only be answered by looking at the client’s real systems:
- Where does the documentation live, what state is it in, who can see what?
- Which ERP data is reliable and which has been contaminated for years?
- Which decisions need a person and which are mechanical?
- Which pilot demonstrates value in weeks, not months?
The blueprint answers this in writing: systems map, architecture diagram, risk matrix, and one or two prioritised pilots with estimated return. It takes a week. And if there’s no project afterwards, you’re still left with a useful document.
Why it’s paid
Because it’s real work, not a disguised sales call. And because a free diagnosis has only one objective: selling you the project. A paid diagnosis has another: being honest — including the possibility that the answer is “not yet”.