External AI officer or an in-house role: how to decide
External or in-house? At the start nobody knows how much work the AI role involves. How to measure it, decide on that basis and hand over cleanly.
In short
An external AI officer pays off as long as it is unclear how much work the role involves in your company. An in-house role pays off once the work arises regularly, you run your own AI projects or a management system under ISO/IEC 42001 is in place. Neither law nor standard prescribes internal or external. Decide on measured work.
Auf Deutsch lesen: deutsche Fassung
“Do we need someone for that?” The question comes up as soon as a company realises that AI has long been part of everyday work. What is usually meant: do we advertise a position? My answer is uncomfortable. At the start nobody knows. Not even you.
I act as external AI officer for several companies. So I look at the question of external or internal from the side that earns its living from it. That is exactly why I also set out here when an in-house role is the better choice and what a handover looks like. An external who makes himself irreplaceable has misunderstood his assignment.
Why the question of a position comes too early at the start
A position needs a description. A description needs an idea of what work will arise. That is exactly what is missing at the start. How many AI tools are running in the company? How often does someone ask whether they may upload a customer drawing to a chatbot? How many approvals are due each month? As long as nobody has counted that, every job description is a guess.
Anyone setting up an AI inventory for the first time often finds more than expected. The assistant in the office suite. The translation function in the mail client. The demand forecast that arrived with the last ERP update. The private chatbot account in sales. None of it was introduced as an AI project. All of it still belongs in the overview, purchased functions just as much as those built in-house.
A position created on suspicion has two possible outcomes. If it is too large, the person looks for work, and rarely the right work. If it is too small, AI becomes a side task for the head of IT, who is fully loaded anyway. The first costs money. The second costs attention, and that attention is then missing exactly where someone is copying customer data into a third-party tool.
What the law and the standard say about filling the role
The short answer: nothing. No law prescribes an AI officer. And no provision requires the role to sit inside your own organisation.
Article 4 of the EU AI Act has applied since 27 July 2026 as amended by Regulation (EU) 2026/1744. Providers and deployers take measures to support the development of their staff’s AI literacy. They do not have to guarantee a particular level for anyone. The article leaves open who organises these measures. Besides a company’s own staff, it expressly mentions persons who work with AI systems on the company’s behalf.
ISO/IEC 42001 is voluntary. Anyone working to it finds three passages on filling the role that together give a clear picture:
- Clause 5.3: top management assigns who ensures that the AI management system conforms to the standard and who reports to it on its performance. Whether that person is an employee is not stated there.
- Clause 7.2, Note 2: as ways of acquiring the necessary competence, the standard lists training, mentoring, reassignment and the hiring or contracting of competent persons. Internal and external stand side by side there.
- Clause 8.1: the organisation must control externally provided processes and services that are relevant to the management system. An external AI officer is such a service. He does not take the control off your hands. You control him.
In data protection, mid-sized companies have known this choice for years. Article 37(6) GDPR expressly allows both: the data protection officer may be a staff member or fulfil the tasks on the basis of a service contract. One difference remains. For data protection, Section 38 of the German Federal Data Protection Act (BDSG) says from what point you have to designate an officer at all: as a rule, from 20 persons who are constantly engaged in the automated processing of personal data. There is no such threshold for AI. You have to find it yourself.
In both models, responsibility remains with the management. An AI officer coordinates and takes work off others. Accountability towards customers, authorities and the workforce stays with the management.
What tasks the role has, and why it makes sense even without a legal obligation, is covered in the basic article:
Does your company need an AI officer?When an external AI officer is the better choice
External fits as long as at least one of these situations applies:
- The work has not yet been measured. Nobody knows whether the role needs four hours a month or two days a week.
- AI is in use, but nobody is officially responsible. The questions land with whoever happens to know a little about it.
- There is nobody in the company who has an overview of tools, the EU AI Act, data protection and standards together and has time for it.
- A customer asks about your handling of AI in a supplier audit. The answer has to be ready within weeks, and recruiting takes longer.
- You first want to find out whether a management system under ISO/IEC 42001 is worthwhile for you at all.
An external brings the view from several companies. Which rule is read on the shop floor and which gathers dust in a folder is quicker to see by comparison. And he has no department to defend. That helps with the question of whether sales may keep its favourite tool.
The downside is part of the picture. I am not on your premises every day. I do not hear what is said about the new tool in the coffee kitchen. A chatbot account created on Monday comes to my attention at the next appointment at the earliest. And I have no authority to give instructions to your people. An external AI officer therefore only works with two things: a person in the company who keeps him informed, and a management that visibly backs his approvals.
In everyday terms, control means: the assignment is in writing before the work begins. The contract with the external should contain at least:
- the scope, in hours or appointments, and who may adjust it
- the authority: may the external block a tool provisionally, or only recommend a block?
- the reporting line to the management, at a fixed rhythm
- confidentiality, and, clarified with your data protection officer, whether the external gets to see personal data
- the handover: all documents belong to the company and are kept in its systems, not on the external’s computer
The last point decides later how easy a change will be. Anyone who negotiates it only at the farewell negotiates from the weaker position.
You want to fill the role externally first and decide after a few months with real figures? That is exactly how my offer as external AI officer is set up, with a handover as soon as an in-house role makes sense.
To the external AI officerHow to tell that an in-house role is worthwhile
An in-house role is worthwhile as soon as the work can no longer be squeezed into appointments. The signs:
- Approval requests come every week, not once a quarter.
- You build your own AI applications or have them built. That can make you a provider within the meaning of the EU AI Act yourself, and the work grows with every project.
- An AI management system under ISO/IEC 42001 is running. Objectives, risk assessment, internal audit, management review and corrective actions need someone who stays on it all year round.
- Customers ask about your handling of AI regularly, not by way of exception.
- There is someone in the team who wants to grow into the role and gets the backing for it.
The last point is the most important. The best internal appointment is rarely a new job advertisement. Often the person is already in the company. The quality manager who knows standards and audits. The production planner who is the first to try every new tool. The head of IT, provided there is spare capacity. What is missing is time and an official mandate from the management.
Three things then belong in the job description. A fixed share of working time, so that AI does not become a side task again. A direct reporting line to the management, as ISO/IEC 42001 provides for in 5.3 b) for reporting on the performance of the management system. And the authority to refuse an approval without having to ask a head of department first.
Between the pure forms lies the mixed model. A person in the company takes on the day-to-day business part-time. The external stays on with a small fixed scope for training, difficult approvals and the view from outside. That takes the best of both sides: closeness inside the company and distance from outside.
Measure instead of estimating: what runs alongside the external phase
The decision between external and internal needs figures. They arise along the way, provided a simple log is kept during the external phase. No software, a spreadsheet is enough. Recorded monthly:
- the number of tools in the AI inventory. If it is still growing, the stocktake is not finished.
- Approval requests with date, tool and outcome.
- Questions from employees, roughly sorted by topic.
- Training: who, when, what content. This is at the same time your evidence of the measures under Article 4 of the EU AI Act and, for those working to ISO/IEC 42001, the documented information on competence under 7.2.
- Customer enquiries about the handling of AI, for example in questionnaires and supplier audits.
- Incidents and near misses: wrong outputs, data in the wrong place, a tool without approval.
- the hours the AI officer actually spends, broken down by these items.
After two quarters you see a curve. If it flattens, a small fixed scope is enough, external or as a part-time task in the company. If it keeps rising, you already hold the job description in your hand. It is in the spreadsheet, line by line.
A thought experiment shows why this holds. Suppose the spreadsheet shows after six months: the inventory is no longer growing, approvals come every few weeks, most questions revolve around the same three tools. Then you do not need a position, you need a good policy and a fixed consultation hour each month. If instead it shows two in-house AI projects in planning, growing customer questionnaires and a management that is thinking about certification, the position is overdue. Both results are good. The only bad one is deciding without a spreadsheet.
That is the real value of the external phase. With it you buy certainty about a question that nobody could answer at the start.
The handover: how to recognise a clean change
If the decision is for an in-house role, the delicate part follows. Knowledge that sits in only one head leaves with the contract. The handover therefore includes at least:
- the AI inventory with status, responsible persons and approval date for each tool
- the AI policy in its current version, including the reasons for contentious rules
- the approval records, including the rejections with reasons
- the training records
- open items and ongoing customer enquiries
- the contacts at the providers of the tools in use
Add to that an overlap in which the new person does the work and the external merely sits alongside. The test at the end is simple. Can the new person answer the next customer question about the handling of AI without calling me? Then the handover is done. If they have to call, it was a dependency and not a service.
One limit applies in both models. Whoever fills the role does not audit it. The internal audit needs objectivity and impartiality (ISO/IEC 42001, 9.2.2 b). Where I am the AI officer of a company, someone else audits there. A certificate is in any case issued exclusively by an accredited certification body.
Measure first. Then fill the role. Then hand over. In that order, the question of a position becomes a decision based on figures.
What you can do on Monday: name a person with whom all AI questions land, in the company or from outside. Have the counting start from that day. In six months you will decide with a spreadsheet instead of a gut feeling.
Primary sources
- Regulation (EU) 2024/1689 (EU AI Act), consolidated version of 27 July 2026, Article 4 AI literacy
- Regulation (EU) 2026/1744 amending the EU AI Act, EUR-Lex
- ISO/IEC 42001:2023, AI management systems, clauses 5.3, 7.2, 8.1 and 9.2.2 (iso.org)
- Regulation (EU) 2016/679 (GDPR), Article 37(6), EUR-Lex
- Section 38 BDSG (German Federal Data Protection Act), data protection officers of private bodies
Five steps to the decision on the AI officer
Name the responsible person
The management determines with whom all AI questions land, in the company or from outside.
Take stock
All AI tools go into the AI inventory, including built-in functions in standard software.
Keep count
Approvals, questions, training, customer enquiries, incidents and the hours spent are recorded monthly.
Decide
After two quarters the curve shows whether a small fixed scope, a mixed model or an in-house role fits.
Hand over
Inventory, policy, records and evidence change hands with an overlap, until the new person works without having to ask.
Frequently asked questions
Is an AI officer required by law?+
No. Neither the EU AI Act nor any German law prescribes an AI officer. Article 4 of the EU AI Act, as amended by Regulation (EU) 2026/1744, requires measures that support the development of staff AI literacy. Who organises that, internally or externally, is for the company to decide.
May the AI officer be an external service provider?+
Yes. No provision requires an internal appointment. ISO/IEC 42001 lists the hiring and the contracting of competent persons side by side in clause 7.2. Anyone working to the standard must control the external service under clause 8.1. Responsibility remains with the management.
From what company size is an in-house role worthwhile?+
There is no fixed threshold, unlike for the data protection officer under Section 38 of the German Federal Data Protection Act (BDSG). What matters is the work that actually arises: approvals, questions, training, in-house AI projects and a running management system. Measure that for a few months before you create a position.
Can the external AI officer also carry out the internal audit?+
No. Whoever fills the role would be auditing their own work. ISO/IEC 42001 requires objectivity and impartiality of the audit process in clause 9.2.2. The internal audit is therefore carried out by another person, and a certificate is issued exclusively by an accredited certification body.
Author & expert review: Lars Zimmermann · ISO/IEC 42001 Senior Lead Auditor & Senior Lead Implementer · ISO/IEC 27001 Lead Auditor & Lead Implementer (PECB)
Last updated: 07 November 2026. Researched and reviewed to the best of our knowledge; not a substitute for individual legal advice.
Questions about your own case?
In a free 15-minute intro call we assess where you stand on ISO 42001, ISO 27001 and the EU AI Act, honestly and without a sales pitch.