Introduction
Welcome to another episode of TJ’s Technology Tuesday and Digital 4 Productivity. Today we are looking at the topic of AI in human resources — and specifically sharing my assessment of a highly interesting article I read in the Whitsun edition of the Handelsblatt. I am a subscriber to the Handelsblatt; I receive no commission whatsoever, but I have to say honestly that I find the quality of the Handelsblatt exceptionally high right now, particularly on the subject of AI. As a supplement, for anyone who has not yet listened to it, I can also highly recommend the podcast by Dr Meckel and the Handelsblatt editor-in-chief, namely Mattes und Meckel.
AI in HR – Where are the limits?
In addition to the podcast, reading remains important, of course. And in the Whitsun edition I came across a very interesting article on the topic of AI in HR. I will display it here in the background in a moment. The title: “How much artificial intelligence can HR handle?” by Julia Bayl. And as I said, the Handelsblatt is generally very much worth reading.
I found this article particularly intelligent because it brought in two aspects that are clearly underrepresented in the current AI debate. What are these two aspects that I believe are always neglected? First, whenever someone says “you can do all of this with AI,” the question to ask is: where exactly are the implementation hurdles? And those hurdles are usually what slows things down. Because when it comes to AI, we generally do not have a knowledge problem — we have an implementation problem.
What additional dimensions exist when deploying AI in HR?
In the field of human resources, there is of course an additional dimension, because we are dealing with highly sensitive, confidential, personal data. In the context of the AI Act, we are very frequently in the domain of high-security AI and high-risk AI. This means there are prohibitions that kick in very quickly. For example, automatically hiring employees without a human reviewing the decision is something that is definitively prohibited under the AI Act. And certain things such as social scoring as well. In other words, this article examines this point and asks: what are the things that may be somewhat more difficult to implement in practice under the AI Act?
And also, drawing on experience with co-determination rights, where might there be situations in which a works council, for example, has a say in the context of co-determination — which, for good reasons, can sometimes slow the pace a little. And that is precisely where I say: here you genuinely cannot operate under the motto “yesterday a Thermomix expert, today an AI expert.” Just as I myself, from my own operational experience as a managing director, have learned first-hand that it is naturally very different whether you do something like this for yourself as a freelancer or whether you introduce it across an entire organisation. So — first point: implementation. Second point, and I also found this very well articulated: everyone is always calling for AI. But the question is, when it comes to payroll, for example, do I actually want a creative answer every time, a creative result each time — or do I want a clearly deterministic, automated, algorithmic process?
Payroll with AI
And I believe that when it comes to payroll, you want a deterministic, cleanly controlled, internally consistent process. And here a conflation often occurs between the topic of automation and the topic of AI. In my own projects it is the same: I am always called in with “Jekel, Jekel, we need AI!” And when we take a closer look, it turns out that around 80% of what is needed is automation and around 20% is genuine AI. And for that I will switch back to the view of the article.
And exactly these points were examined very carefully by the author, in my view. She said: OK, so much HR is involved in AI — and she then looked at the top section, HR administration, and differentiated between the digital automation potential on the one hand and the AI potential on the other. So these two columns side by side — I find this a very, very good approach because it is a very differentiated view of the topic. You can see very clearly that in HR administration, for example, there is a very high automation potential of 40 to 55% of activities that can be automated. But here we are precisely in the domain of topics like payroll.
Unstructured processes can be very well supported by AI
And there we do not want to be creative. And of course there is also a potential AI topic here — for example, answering employee questions with chatbots. And here you can clearly see the distinction, where the boundary between automation and AI becomes a little more visible. Automation applies whenever you have structured, clear, deterministic processes — such as payroll. AI applies whenever you have unstructured processes or unstructured inputs to deal with.
If you have a chatbot for employee questions, for instance, you used to create Q&As. But then someone phrased the question differently from how it was programmed in the Q&A session, and no sensible answer came out. AI can take many differently formulated questions and work out: what does this employee actually mean?
A blend of sensible automation and AI supplementation?
Exactly — and in that context, you match the question against the answers you have defined once, or you match it — without having defined specific questions — against the content you have stored. I recommend, by the way, that even when building chatbots with document access where the AI retrieves information, you should always say: step one is Q&As, and only if the answer is not found there does it go into the documents. So this combination is not either/or — it is always a blend of sensible automation and sensible AI supplementation. And upstream of all that, of course, the fundamental question: do we actually have a problem here? Because I see many AI tools and many AI automations where I say — please tell me what problem you are solving with this.
AI is strong at recognising patterns
And I see in many organisations that IT increases complexity, and then a thousand things get done. There was also an interesting Handelsblatt article in the Whitsun edition about how to re-engage employees who are simply wasting hours with AI because they are not doing value-adding activities with it. That is precisely the mistake, and that is why I find this article so good — because it has this focus. Looking again at HR controlling, you can see there is more potential for AI here, which makes sense, because this is where you have people analytics, where you have more identification of patterns — and AI is strong at recognising patterns. Also here in the topic of HR controlling: what is pattern recognition?
It is statistics. And ultimately AI is, to a certain degree, statistics on steroids — statistics, but genuinely simple to use and actually actionable. Against this background, it helps enormously to have engaged with the topic of statistics at some point. I am very grateful that in my business administration degree, in my doctoral studies and in my Executive MBA I really did engage intensively with statistics. My wife even added a PhD on top, which means I also had the pleasure of accompanying that journey and thus gained a certain feel for how statistics works.
AI as a multiplier
And AI is a multiplier — which means that if you have no clear idea and your output is below 100%, i.e. you have a factor of less than 1, and you multiply that with AI, you will get less out of it than you would from AI alone. So it is always a multiplier that you need to keep in mind. Let us continue through the points in this article. The topic of compensation and benefits: here we are again somewhat higher in terms of automation potential and somewhat lower in terms of AI potential.
Artificial Intelligence or Automation?
And here again: please do not be misled by these marketing claims about AI. I remember attending the last Zukunft Personal conference, walking through the halls and asking at every stand where AI was mentioned: “Tell me, what exactly are you doing with AI?” And then some of them started to stammer quite badly. “Well, we just sort of wrote it down.” And very often, when I did get an answer, I said: “Is that AI or is that automation?”
Did you not already have that before? In other words, you always have to say clearly: productivity is always beyond the hype — beyond all the fanfare, the hullabaloo and “this will save the world.” You know the Hype Cycle: at the beginning we say yes, everything is fantastic; then comes the trough of disillusionment, where we say everything is rubbish and nothing works. Those who are productive move into a zone that is below the inflated expectations but above the trough of disillusionment, where many unsuccessful companies also disappear. They climb.
AI and Recruiting – What biases exist?
Let us look further. On the topic of recruiting, it is very clear that you have to be extraordinarily careful — on one hand in terms of regulation, and on the other hand in terms of bias. You may have followed the examples from Amazon and Microsoft, for instance, who tried to use AI for recruiting. And the most vivid example for me is always the one from our former Federal Minister of Labour and Social Affairs, Hubertus Heil, who once told Markus Lanz that they had “built the handbrake using AI.” What was this handbrake?
The handbrake, very simply: they had tried to use an AI to optimise recruiting for state secretaries — i.e. to evaluate applications. The problem was that the training data consisted of all successful state secretaries, and they were all male and all named Hans. So that was the so-called handbrake, because anyone who was male and named Hans immediately moved up in statistical probability. Or Amazon and Microsoft, who in the past had predominantly hired men for software development positions. So of course what happened?
Yes — if you are male, you have a higher probability of succeeding there. Which is statistically correct, but which is clearly wrong from a diversity perspective. And that is why: always switch your brain on first, then the technology — especially in the HR domain. For the topic of strategic workforce planning, as a supplement in the areas of personnel development, training, and employee support. This is, as I said, a very, very good article in the Handelsblatt that makes clear it is not about AI replacing HR work — it is also not purely AI, but rather automation, and supplementing that with what matters in many other areas too: the human side of business.
When does AI in HR make sense?
Because the AI Act states that neither personnel decisions nor credit decisions may be made 100% automatically, without a human reviewing them first. And it is a good idea to have human judgment involved, particularly for processes that do not need to be decided in milliseconds. Let me give a counter-example: looking at the Iron Dome in Israel, the rockets coming from neighbouring countries simply have such short lead times that AI systems have to decide very, very quickly — friend or foe, aircraft or missile — and in case of doubt must launch the interceptor missile as part of a defensive strategy. But hiring staff has nothing to do with warfare, hopefully — it has to do with strategic thinking. And that is precisely where the concept of a human in the loop is absolutely essential.
Conclusion
And additionally, of course: understanding the human factor, bringing the human factor into the equation. The good news is that when AI and automation help us create space — by automating routine tasks and supporting decision-makers with better data — giving us more personal freedom for strategic decisions, for the strategic care of the most important resource in any company, namely the people, the employees — then AI makes sense.
If you need support in these or other areas as a sparring partner at C-level, please feel free to contact me as a Personal IT Coach for executives.
And on that note: switch your brain on first, then the technology. Yours, Thorsten Jekel.
Key Takeaways
- AI in HR is not a cure-all – the Handelsblatt article provides a differentiated view of where AI is genuinely useful and where automation or human judgment is the right answer.
- Automation and AI are two distinct concepts: structured processes such as payroll are suited to automation, while unstructured processes such as answering employee questions benefit from AI.
- In practice, according to Thorsten Jekel, roughly 80% of solutions are automation and only 20% genuine AI – even though companies typically call for “AI.”
- The HR domain is subject to particularly strict rules: the EU AI Act classifies many HR applications as high-risk AI and prohibits, for example, fully automated hiring decisions without human oversight.
- Bias in recruiting is a real risk: well-known examples from Amazon, Microsoft and the German Federal Ministry show how training data can amplify discriminatory patterns.
- AI is particularly strong at pattern recognition (people analytics, HR controlling) – which is essentially statistics taken to a higher level.
- AI acts as a multiplier: those who start without solid expertise or a clear problem definition gain no added value from AI – it amplifies weaknesses rather than strengths.
- The principle “switch your brain on first, then the technology” applies especially in HR, since decisions affect people and require strategic thinking.
- Human in the Loop is mandatory: both ethically and legally (AI Act), HR decisions must always be reviewed by a human being.
- Sensible AI deployment creates space for strategic HR work and the care of the most important resource in any organisation: the people.
Frequently Asked Questions
What is the difference between AI and automation in HR?
Automation is suited to structured, deterministic processes such as payroll, where the same clear result is always expected. Artificial intelligence is used where unstructured inputs or processes are involved – for example, answering differently formulated employee questions via a chatbot. In practice, a blend of both approaches is usually the best solution.
Is AI allowed to make hiring decisions autonomously in HR?
No. The EU AI Act classifies fully automated hiring decisions without human oversight as prohibited, since they constitute high-risk AI. A human must always review the decision before it takes effect.
What risks does AI pose in recruiting?
AI systems can absorb and amplify bias from training data. Amazon and Microsoft, for example, favoured male candidates when using AI for software development roles because historical hiring data was predominantly male. A further example is the German Federal Ministry, which attempted to use AI for selecting state secretaries and found that the model favoured candidates named “Hans.”
Why is payroll not a good use case for generative AI?
In payroll, a deterministic, consistent and traceable result is required – creative or variable outputs would be a disadvantage here. Such structured, rule-based processes are better suited to classical automation than to generative AI.
Where does AI have the greatest potential in HR?
AI unfolds its greatest potential in HR through pattern recognition – for example in HR controlling and people analytics. It is also considered a valuable addition for answering employee questions via chatbot, for strategic workforce planning and for personnel development.
What does “Human in the Loop” mean for AI-supported HR decisions?
“Human in the Loop” means that for decisions affecting people – such as hiring or promotion – a human always reviews the AI recommendation and bears final responsibility for the decision. This is both an ethical requirement and a legal obligation under the EU AI Act.
How does AI act as a multiplier in organisations?
AI amplifies what is already in place: those with solid expertise and clearly defined processes benefit enormously from AI. Those who start without a clear problem definition or professional foundation gain no added value – the multiplier effect then falls below 1, meaning it has a negative impact.
How should an AI chatbot for employee questions be built?
A two-tier architecture is recommended: the chatbot first searches predefined Q&A pairs; only if no answer is found there does it fall back on stored documents. This combination of structured answers and document-based AI search delivers more reliable results than a purely document-based approach.
Why do many AI projects in HR fail at the implementation stage?
The biggest obstacle is rarely a lack of knowledge, but the implementation itself: high regulatory requirements from the AI Act, sensitive personal data, co-determination rights of works councils and the absence of a clear problem definition all slow many projects down. Added to this is the fact that vendors often promise “AI” when what they deliver is simple automation.
When does deploying AI in HR genuinely make sense?
AI makes sense in HR when it creates space: by taking over routine tasks and supplying decision-makers with better data, so that HR professionals can focus on what matters most – the strategic care and development of employees as the most important resource in any organisation.
Tools & Resources Mentioned
- AI overview on digital4productivity.de – Further content on artificial intelligence in a business context
- Handelsblatt – Source of the article discussed: “How much artificial intelligence can HR handle?” by Julia Bayl (Whitsun edition)
- Podcast “Mattes und Meckel” – Recommended podcast by Dr Meckel and the Handelsblatt editor-in-chief on the topic of AI
