OWNERSHIP · 2–4 DAYS A WEEK · MONTHLY RETAINER
Fractional Chief AI Officer, in the room when the decisions get made.
Two to four days a week of senior ownership over your AI direction, without carrying a full-time executive at that level on the payroll.
THE PROBLEM THIS SOLVES
The problem this solves
Companies between twenty and five hundred people reach a point where AI decisions stop being experiments and start having consequences. A vendor contract gets signed. A data pipeline gets built on assumptions nobody wrote down. A department starts depending on a workflow that one person set up in an afternoon.
Nobody owns those decisions. The CTO owns delivery, the COO owns operations, and AI sits in the space between them where it belongs to everyone and therefore to no one. Hiring a full-time Chief AI Officer at that headcount is difficult to justify to a board, and the search itself takes six months if it succeeds at all.
WHAT THE ROLE COVERS
What the role covers
- Owns the AI roadmap, including the sequence in which things get built and what gets refused
- Attends the leadership meeting as the accountable voice on AI decisions, not as a guest
- Evaluates vendors and reads the contracts, including the clauses about your data
- Sets internal governance: who may deploy what, on which data, with which review, mapped against the obligations that actually apply to your company
- Supervises the internal engineers or outside agencies doing the build
- Writes a monthly brief the leadership team can act on, with named decisions and named owners
HOW IT WORKS
How it works
Two to four days a week, or eight to sixteen days a month if that suits your calendar better than a fixed weekday. Three months minimum, because anything shorter produces opinions rather than change. After the first three months the arrangement continues month to month and either side can end it with thirty days' notice.
Work happens on site in Munich and the wider DACH region for leadership meetings, and remotely for everything else.
WHAT THIS IS NOT
What this is not
This is not engineering capacity and it is not a training programme. It is also not a monthly report that arrives and gets filed. The arrangement only works if the person is in the room when the decisions get made and has the standing to make them.
- vs an AI consultant
- A consultant produces a recommendation and leaves. The accountability for acting on it stays with you, which is why so many good recommendations are still sitting in a folder. This role holds the decision instead: it owns the sequence in which things get built and what gets refused, and it attends the leadership meeting as the accountable voice rather than as a guest. If what you need is a diagnosis rather than an owner, the AI readiness assessment is the cheaper and more honest answer.
- vs a full-time hire
- The same seniority, two to four days a week instead of five, with a three-month minimum and thirty days' notice after that. The trade is real and worth stating: fewer days means the time has to go on decisions and governance rather than on delivery, so this works where the bottleneck is direction and ownership, and does not where you need someone building.
WHAT THE DECISION IS WORTH
The constraint was chosen first.
This is what the role is for, and the clearest way to show it is two calls that were already made. In each, the decision came before the work started and set what the work could cost — the part no amount of good engineering afterwards recovers. Neither is a system, which is why neither is in the portfolio: the output was a decision, and what it ruled out.
Treating retraining cost as the constraint, not the consequence
At an enterprise software company, 300 client companies and roughly 15,000 end users moved from a Windows desktop application to the browser over 18 months. No end-user retraining was needed, because the interface was deliberately preserved.
The engineering was considerable. The decision came first, and it set what the engineering had to achieve: end-user retraining cost was named the project's primary constraint before any code was written, and every technical choice after that was made subordinate to it. Preserving the interface was not a nice outcome that emerged — it was the requirement the rest of the design had to satisfy.
What the users saw did not change.
Naming that constraint up front is the whole of it. Named afterwards, it becomes a change-management problem for somebody else, on a scale of 15,000 people.
Renting inference rather than buying the server
A dedicated on-premise inference server was on the table. The hardware quote was €100,000, carrying an 18-month warranty and no manufacturer support once it expired. The alternative was renting the same inference at €3,000 a month, with the provider carrying the cost of replacing hardware as it aged.
Across the warranty window, renting came to roughly half. That is the smaller half of the argument. The larger half is what each option leaves you holding at the end of those 18 months: one leaves you free to decide again, the other leaves an unsupported box in production — in a market where inference hardware ages out in 18 to 24 months.
Recorded as a business case: expenditure not incurred. Not a realised saving, and not presented as one.
FIT
Fit
This fits when
- AI has moved past a pilot and something is now in production, or about to be
- There is budget for AI work but no single person accountable for where it goes
- Leadership wants one owner rather than three vendors with three roadmaps
- A full-time hire at this level is either unaffordable or unfillable in your market
This does not fit when
- What you need is someone to write the code
- You already have a full-time AI lead and want a second opinion — that is an AI Workflow Audit, not this
- You are under twenty people and one founder can still hold the whole picture
FEE
Fee
A flat monthly retainer, quoted after a thirty-minute conversation. It depends on the number of days and on how much of the roadmap you want owned rather than advised on.
The retainer is optional and it changes nothing about ownership. Everything built during the arrangement stays with your team, the same as in every other engagement here.
FAQ
Common questions.
How is this different from the AI Workflow Audit?
The Audit is a two-week diagnosis with a defined scope and an end date. It tells you where the constraint sits. This is ongoing ownership of what you do about it. Several companies start with the Audit and only then know whether the role is worth filling.
Do I have to commit beyond three months?
No. Three months is the minimum because the first month is spent understanding how the company actually works rather than how the org chart says it works. After that it runs month to month with thirty days' notice.
Can this work alongside our existing CTO?
That is the normal case. The CTO owns delivery and the engineering organisation. This role owns AI direction and the decisions that cross departments, which is precisely the set of decisions a CTO has no time to own and no mandate to force.
Is the work on site or remote?
Both. Leadership meetings in Munich and the DACH region are attended in person. The rest is remote, which is what makes the arrangement affordable rather than the same days plus travel.
What has to exist by the end of the first three months?
A roadmap with a stated sequence and a stated list of what is being refused. Written governance covering who may deploy what, on which data, and with which review. A position on the vendors and contracts already in play. And the monthly brief running, with named decisions and named owners. Three months is the minimum precisely because the first of them goes on understanding how the company actually works rather than how the org chart says it works. This is a set of outcomes, not a week-by-week plan — the sequence depends on what the first month finds.
Who is responsible for EU AI Act obligations?
Day to day, this role. Before anything is built, it forces the question that decides everything downstream: does this system fall under the Act at all, and if it does, what does the Act require it to have? Once that is settled, the role keeps three things true. The governance written down matches the obligations that actually apply to your company. Human oversight is proportionate to the risk rather than uniform across everything. And every obligation has a named owner instead of a shared assumption that somebody is handling it. What this role does not do is sign the legal determination. Whether a given system is high risk is your counsel's call, and anyone offering to make that call for you is selling an opinion as cover. The split is the ordinary one: your lawyer rules on what applies, and this role makes sure the engineering and the governance actually do it.
Does this role decide build versus buy?
It owns the decision and the reasoning behind it. That means evaluating the vendors properly, reading the contracts including the clauses about your data, and being able to say why a thing is being bought rather than built — or refused entirely. The most valuable output of the role is often the refusal, because the cost of the wrong platform is paid over years and lands on somebody who did not choose it.
How does the leadership team stay informed?
Two ways, and both are part of the arrangement rather than extras. The leadership meeting is attended in person as the accountable voice on AI decisions. And a monthly brief goes to the leadership team with named decisions and named owners — written to be acted on, which is the difference between a brief and a status report.