Blueprint: hiring a Chief AI Officer (what the market asks for, what it leaves unsaid and what that person must build)

If your board has decided to "hire a Chief AI Officer", start with an uncomfortable fact: across 79 AI leadership postings published between November 2025 and July 2026 in Switzerland, DACH, Benelux, the Nordics and France, none uses that title. Not one. What the market publishes is 24 titles beginning with Head of AI, Head of Artificial Intelligence or Head of Data & AI, and the rest are transformation, adoption, program or function leads (AI for finance, for operations, for marketing). The fashionable title lives at conferences; the work lives in postings with humbler names and more operational duties.

Read closely, those postings describe a building role, not boardroom-deck strategy: 13 of 79 explicitly ask the hire to build the team from scratch, 10 require the person to stay hands-on, and 32 hang company-wide training and change management on them. The practical conclusion for you: you are not hiring a title, you are hiring the construction of an AI operating system for your company. This blueprint is ordered accordingly: first the choice of figure, then what you must demand in writing, then what that person must build and with whom. The full method we use to run our own systems is documented in the reference case and the content operation blueprint.

The most expensive mistake is not overpaying: it is hiring the wrong figure for your phase. Of the 33 postings in the corpus that state a reporting line, only 15 hang from the C-suite; the rest report to functional directors, which already tells you most European companies are buying execution, not a seat at the table. The table sums up the five real figures in the market with the rule for when each one makes sense. The usual trap: asking for a C-suite profile (vision, board, regulation) and evaluating with a technician's checklist (frameworks, code, tool certifications). Decide first what you are buying; write the posting after.

The next table is the corpus talking: each row is a signal counted across the 79 postings, with what it means for you as an employer. The underlying pattern: companies already know they need AI governance (43% put it in the role's duties), but almost none can yet name the concrete legal framework that applies to them. Only one posting out of 79 names the EU AI Act, and of the 21 Swiss postings none mentions the nFADP. That is an opportunity if you hire well: the person who can land those frameworks puts you ahead of your sector.

Then there is what does not get written. The same postings that ask for "strategic vision" describe builder duties: assemble the team (13 of 79), stay hands-on (10), train the whole workforce (32) and answer for measurable results (38 demand measurable KPIs, though only 4 dare to state a numeric target). The tacit requirement is a triple profile: can build systems, can govern them and can win over a skeptical organization. When you interview, test the three separately; the candidate brilliant at only one will fail you on the other two.

Whichever figure you hire, the deliverable is the same: an AI operating system with four layers feeding each other in cascade. Layer 1 sets the frame (what is allowed, at what risk, with what budget); layer 2 picks and pilots use cases inside that frame; layer 3 provides the platform and data that make them possible; layer 4 makes it last: trains, measures and reports to leadership. The diagram's reading rule: nothing in layer 2 launches without layer 1's frame, nothing in layer 3 gets bought without a layer 2 case justifying it, and layer 4 decides every quarter what scales and what gets killed. If your candidate cannot draw you this system (theirs, not ours) on a whiteboard, keep interviewing.

This is the plan you should agree with the person before they sign, because it is also your checklist as the employer: every step needs something from you (access, sponsorship, budget). If at 90 days there is no inventory, no policy, no two measured pilots and no working committee, the problem is the plan or the sponsorship, not necessarily the person.

Of the 79 postings, 13 explicitly ask the hire to build the team from scratch and only 5 dare to state a size (between 5 and 17 people). The rest stay silent, and that silence is the hidden cost of the hire: a Head of AI with no follow-on team becomes an expensive bottleneck who does everything and scales nothing. The table gives the realistic order of arrival. The rule: each hire is justified by a pilot that already proved value and was limited by the lack of that exact capability, not by an aspirational org chart. In an SMB the first rows can be the same person or part-time profiles; what matters is the order, not the headcount.

The corpus captures the market's confusion: 38 of 79 postings demand measurable KPIs and only 4 commit to a numeric target. Translation: everyone wants accountability and almost nobody knows yet what account to ask for. Do not ask for "AI ROI" as a single number in year one; ask for the table's frame, which produces defensible numbers and detects theater. The detail of how to measure return without inflating it is in how to measure AI ROI: the mechanics are identical outside marketing.

Each table row is a failure mode we have seen or operated close to, with its early signal: what a board can watch without being technical. The corpus fact that should worry you most sits in the second row: 61 of 79 postings name no concrete AI vendor, which sounds like neutrality but usually means the dependency decision will be made later, with no written criteria, by way of enthusiasm.

So you can trust the numbers above, here is what they are not: they are not a random sample of the market. The corpus holds 79 in-scope postings (of 83 collected), captured on July 24, 2026 over a November 2025 to July 2026 window, from company career pages, public ATS systems and job boards; it over-represents employers whose ATS pages are publicly crawlable and under-represents the Swiss SMB that only posts on closed portals. 74 of the 79 rows are full-text extractions of the posting; 5 are snippet-level and marked as such. Every figure cited here comes from frozen definitions independently audited (recounted datum by datum before publication), and every dataset row keeps the URL where the posting was seen.

This is the pilot corpus of a larger study: the full version will grow the sample to 150-300 postings under the same definitions, and will be published with the open, auditable dataset. If the numbers move as the sample grows, we will say so right here, with a date.

Three shifts already read in the corpus. First, from advising to operating: 20 of 79 postings already mention agents in the duties, and we expect that fraction to grow; the AI leader of 2027 will run fleets of agents with a budget and a failure log, not decks about potential. Second, governance moves from virtue to obligation with teeth: the AI Act's obligations apply in phases, and the gap between the 43% asking for generic governance and the 1% able to name the law will close, willingly or by sanction. Third, the title matters less and less and the system more and more: the market converges on Head of AI with a cross-functional mandate in mid-sized companies and on the fractional model in SMBs (one in four corpus postings is already advisory-side). Our honest read: if you wait for the perfect title, you will be late to the system; if you build the four-layer system this year, whatever you call it afterwards will hardly matter.

The person signing this operates their own systems on this same scheme, with costs and failures published. If you want the version applied to your company, the door is below.