career bridge

The inventory

You cannot govern what you cannot see. Six fields per system, three independent sources, ranked by consequence rather than spend.

You know which rules reach the organisation. Now the uncomfortable part: nobody can tell you what AI is already running inside it. Procurement knows about two systems, three teams are quietly using a fourth, and somebody in marketing has a subscription no one has ever reviewed.

What you do

Go and find it, because you cannot govern what you cannot see. Build the inventory from more than one direction, since any single source will miss things. Ask teams directly and expect incomplete answers, not because people lie but because a feature inside a tool they already had does not feel like adopting AI. Read the expense and subscription records, which is usually where the surprises are. Check what the systems you already own now ship with, because vendors keep adding these features and enabling them by default. For each thing you find, record six fields and no more: what it does, who owns it, what data goes in, whether it makes or influences a decision about a person, whether a human reviews that decision, and where it runs. Resist a longer form. A thirty-field template returns nothing because nobody completes it, and a six-field inventory that is actually finished beats a beautiful one covering a third of the estate. Then rank by consequence rather than by spend, because the riskiest system is rarely the most expensive one.

Done when

  • The inventory is built from at least three independent sources, not one survey.
  • Every entry records what it does, its owner, its data, whether it decides about people, human review, and where it runs.
  • At least one system appears that nobody in procurement knew about.
  • Entries are ranked by consequence rather than by cost.
  • The form is short enough that an owner can complete an entry in ten minutes.

What you end up with

An AI inventory for a real organisation: six fields per system, ranked by consequence, including at least one system nobody had recorded.

If you get stuck

The common failure is a long questionnaire that comes back empty, which then gets reported upward as low adoption. Low response is not low adoption. Shorten the form and go to the expense records instead. The other failure is treating this as a one-off census. Put a date on it and a review interval, because the estate will have changed before you finish writing it up.