Skip to content
caadoo
de

Fractional Chief AI Officer: what the role covers

Large companies hire a Chief AI Officer. A company of 20 to 250 people rarely needs one full-time, but somebody still has to own the job. This piece covers what that job contains, the signs it's missing, and when a fractional CAIO is the wrong answer.

By T. Krause12 min read

A fractional Chief AI Officer is an outside person who takes responsibility for AI in your company for a fixed share of their time. Inside an agreed scope, they decide which tools are allowed and which project comes next. Afterwards they measure the result. The role is meant for companies of 20 to 250 people, where the job rarely fills a full-time position. Some call the same role an external Chief AI Officer.

Ask three people in your company who is responsible for AI and you'll probably get three different names, or a shrug. Sales pays for a ChatGPT subscription on somebody's personal card. The accounting software started reading receipts after its last update, and nobody remembers agreeing to that. A team lead built a small automation that pushes order confirmations into the ERP, and she is the only one who knows how it works.

None of that is a problem on its own. Taken together, it's a set of decisions nobody made on purpose. No one can say which data sits in which account or who has been trained. Large companies call the person who sorts this out a Chief AI Officer. This piece is about the same job in a smaller company.

I sell a version of this myself, so read with that in mind. It's also why there's a section further down on when a fractional CAIO is the wrong call.

Photo: Unsplash

What the job contains at 20 to 250 people

Strip away the title and what's left is ownership. Research and custom models aren't part of it.

The process inventory. Before any tool decision, you need a list of where the hours actually go. The fastest honest version is a tally sheet each team keeps for a week, as described in the piece on choosing your first automation.

Priorities. The list becomes an order. Frequent, dull, measurable tasks go first. Somebody has to hold that line when a department head comes back from a trade fair with a favourite project, and that only works if holding it is their job.

Tooling decisions. Which company accounts exist, which data may go into which tool, who has admin rights, and what it all costs per month. That includes whether to switch on the AI features that turn up uninvited in software you already pay for.

AI literacy under Article 4. The EU AI Act requires providers and deployers of AI systems to take measures that support AI literacy among their staff. Regulation (EU) 2026/1744, known as the Digital Omnibus, rewrote the article, and it no longer asks for a specific or "sufficient" level. The European Commission's AI literacy Q&A says national market surveillance authorities supervise and enforce the duty from 2 August 2026. The same Q&A says Article 4 mandates no particular governance structure, so the law doesn't require you to appoint a Chief AI Officer. Somebody still has to plan the training for each role and keep the record. The details are in the piece on what Article 4 asks of companies.

The works council. If you have a Betriebsrat, German law gives it a say early. Section 90(1) no. 3 of the Works Constitution Act (BetrVG) requires the employer to inform it in good time about planned work processes, and the text names the use of AI explicitly. Tools that can monitor employees' behaviour or performance are subject to co-determination under section 87(1) no. 6. Talking to the works council before the purchase order costs an afternoon. Talking to them afterwards can cost months.

Data protection. Every tool that processes personal data needs a processing agreement under Article 28 GDPR. Processing that is likely to pose a high risk to the people affected also needs an impact assessment under Article 35. The CAIO doesn't replace your data protection officer. Under section 38 of the German Federal Data Protection Act (BDSG), you need one where, as a rule, at least 20 people are constantly engaged in the automated processing of personal data. Keep the two roles apart. The person who picks a tool shouldn't be the one who signs it off.

Measurement. A baseline number before each build, a second number four weeks after go-live. Skip the first one and you can't prove anything later, which is what the piece on taking a baseline first is about.

Runbooks. One page per automation. What it does, where it runs, who gets called when it breaks, how to switch it off. Without that page the knowledge belongs to whoever built it and leaves when they do. There's a template in the piece on automations that survive staff turnover.

Signs the role is missing

None of these is a crisis. They cost time and money that never show up on a budget line.

  • Nobody can produce a list of the AI tools in use, only guesses.
  • Staff paste customer data into personal accounts because there's no company account, or nobody has said what's allowed.
  • Three departments pay for three overlapping subscriptions.
  • A pilot has been running for months and nobody can say whether it paid off.
  • The one automation that really works depends on one person.
  • If a regulator or a customer asked who has been trained on AI, you couldn't show them.
  • AI is on the leadership agenda every month, and nobody leaves the meeting with a task.

If several of these sound familiar, another tool won't fix it. What's missing is an owner.

In-house hire, fractional or agency

Hiring in-house. The person is there every day, knows the people, and what they learn stays in the company. The hard part is finding one person who covers processes, tooling, regulation and training, and then giving them enough of that work. In a company of 40, the AI part rarely fills a week. The role drifts into general IT support and the AI work waits. Hiring makes most sense when someone on staff already does part of the job and you can clear time for them.

Fractional. An outside person holds the role for a fixed share of their time, usually with a standing weekly slot and an agreed scope. You get someone who has worked through the same questions at other companies, without creating a full position. The cost is presence. They aren't in the building every day, and they start out knowing nothing about how you work. Without an internal counterpart who opens doors and pushes decisions through, a fractional CAIO turns into an adviser with a calendar.

Agency projects. An agency builds a defined thing, say an invoice workflow or a customer chatbot, at a fixed price or on time and materials. For a well-scoped project that's a sound route. It doesn't fill the role. When the project ends, the agency moves on, and the questions about what comes next and who keeps it running land back on your desk. Agencies also make their money by building, which makes them a poor referee on whether something should be built at all. How a project, a platform and a subscription compare on cost is covered in what AI automation costs a small business.

When a fractional CAIO is the wrong call

AI is your product. If you build software with AI at its core, that knowledge belongs in your engineering team. Hire.

Nobody inside has time. A fractional CAIO can prepare decisions and make the ones inside their scope. They can't grant themselves access or chase colleagues down the corridor. If nobody internal can give a few hours a week to answering questions and signing things off, the outside hours are wasted.

You need exactly one project. One automation and nothing planned after it? Book a scoped project. A standing role is overhead you don't need.

The work is on the shop floor. Changing production processes day to day takes someone who is physically there. A weekly video call won't do it.

Leadership isn't behind it. An outside person can't enforce priorities the managing director won't back. Settle that first.

What to agree for the first 30 and 90 days

Whoever holds the role, write down beforehand what should exist at day 30 and at day 90. The first review then checks a list instead of trading impressions.

By day 30.

  • A named internal counterpart with fixed hours each week.
  • An inventory of every AI tool in use, with purpose, department, account type and the kind of data that goes in. Honest answers come from an open survey with an amnesty, not from trawling logs. The piece on shadow AI in the workplace has the questions to ask.
  • A week of tally sheets from the teams that do the most retyping, and from that a shortlist of three to five candidates.
  • A baseline number for the first candidate.
  • A first conversation with the works council, if you have one, before anything is bought.
  • A provisional rule on which data may go into which tool.

By day 90.

  • The first automation is live, with a before number and an after number.
  • Every role has had its training, and a table records who got what and when.
  • A usage policy that people actually know about. The AI usage policy template suggests half an hour with each team, walking through two examples from its own work.
  • A runbook for every automation that's running.
  • A priority list for the next quarter, including what was dropped and why.

For the day you part ways. Accounts, credentials, automations and documentation all live in your company's accounts, never in the outside person's. Put that in the contract before the first build. Whoever leaves hands over a list, and everything keeps running.

The title is big, and the job behind it fits on a page. Someone knows what's running and decides what comes next. Whether that's a new hire, a fractional CAIO or the managing director with Friday mornings blocked out matters less than having a name next to it.

caadoo offers this role as a subscription for companies of 20 to 250 people. The plans page shows what each plan covers.

Frequently asked questions

What does a fractional Chief AI Officer do?

An outside person takes ownership of AI in your company for a fixed share of their time. They keep the list of processes where hours go missing and set the order of work. They settle which data may go into which tool and plan the AI literacy training. For every automation they take a number before and after and write a runbook. They make decisions inside an agreed scope. Research and custom models aren't part of the job, and the role doesn't replace your data protection officer.

Should you hire a Chief AI Officer or use a fractional one?

Hire if AI is your product. If production processes change day to day on the shop floor, you need someone on site. Hiring also makes sense when someone on staff already does part of the job and you can clear time for them. Fractional fits when the AI work doesn't fill a full position and someone internal has a few hours a week for questions and sign-offs. Leadership has to back the priorities. For exactly one project, a fixed-scope contract costs less than a standing role.

At what company size does a Chief AI Officer make sense?

I won't name a threshold, because I can't back one up. This piece and caadoo's subscription are written for companies of 20 to 250 people. In a company of 40, the AI part rarely fills a week, and somebody still has to own it. The signs in the section above tell you more than headcount, for example that nobody can produce a list of the AI tools in use, only guesses. If several of them sound familiar, what's missing is an owner.

What happens in the first 90 days with a fractional Chief AI Officer?

Write down beforehand what should exist at day 30 and at day 90. By day 30 there should be a named internal counterpart with fixed hours each week and an inventory of every AI tool in use. By day 90 the first automation should be live, with a before number and an after number. Every role has had its training by then, written down in a training record. People know the AI usage policy. The full lists are further up and apply to an in-house hire too.

How much does a fractional Chief AI Officer cost?

It depends on the agreed scope, meaning the share of their time the person holds the role for. caadoo offers the role as a subscription at a fixed monthly price. The plans page shows what the plans cost and cover. Tool licenses and API costs are not included there and run in your company's name. How a project, a platform and a subscription compare on cost is covered in what AI automation costs a small business.

Does the EU AI Act require a Chief AI Officer?

No, the EU AI Act doesn't require you to appoint a Chief AI Officer. The European Commission's AI literacy Q&A says no specific governance structure is mandated to comply with Article 4. It says Article 4 requires providers and deployers of AI systems to take measures to support the development of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf. Since the Digital Omnibus, Regulation (EU) 2026/1744, no specific or "sufficient" level is mandated. The obligation remains.

Topics

  • Chief AI Officer
  • AI strategy
  • Mittelstand
  • EU AI Act
  • Automation
All articles

Keep reading

  • 8 min read

    Automations that survive the builder leaving

    Some automations run only because the one person who built them is still around. That works until they are on holiday, off sick or gone. This is how to take stock of what already runs and give every automation an owner, a one-page runbook, a team alert address, a shared vault and a quarterly check.

  • 7 min read

    Measure before you build: the baseline nobody takes

    Most automations are called a success because the work feels faster afterwards. Nobody can say how much faster, because nobody measured before. This is how to take a two-week baseline and write the one-line before and after statement a managing director can check.

End of call

You can hang up now.

Or book 15 minutes. I'll ask you the questions I'd ask on the phone, and tell you at the end whether it fits. If it doesn't, we've both lost a quarter of an hour instead of a month.

Or call. Cold, if you like. My AI assistant picks up and I call you back. +49 5136 9414999

Analytics

With your consent I use Google Analytics to see which pages get read. Without it, no analytics code loads. Your answer is stored locally in your browser.

Privacy notice