Most AI projects we get called into have already bought something. A licence, a pilot, a platform. Then the questions start, and every one of them is about plumbing.

Where does the data live. Who is allowed to see it. Is it structured enough to query, or is it four years of PDFs in a shared drive. If the model gets something wrong, who notices. And when someone leaves the company, how do you take the access away.

Answering those is the actual project. The tool is the easy part and it should be the last decision, not the first.

Being ready looks like this.

None of that is AI work. It is network, cloud and security work, which is why the companies getting value from AI fastest tend to be the ones whose systems were already in decent shape.

The failure we see most often is a model answering confidently from the wrong version of a document, because three versions exist and nobody could say which was current. That was never an AI problem. It would have bitten you eventually anyway.

If you want to know where you stand, try this. Pick the single question you most want to ask an AI tool. Then trace by hand where the answer would have to come from, system by system, permission by permission. Whatever that exercise turns up is the work, and you will usually find it before you finish the first system.

Find out what your systems are ready for ›››

All posts ›››