AI in HR: automate processes, not humanity

AI belongs in people work, but in the right place. Today it is excellent at taking over repetitive work: drafts, summaries, process logistics, access to knowledge. It has no business in the moments where a person sits across from a person: feedback, conflict, separations, final hiring decisions, and the moments when someone simply needs to be heard. What matters is a clean dividing line and an introduction with clear guard rails.

Where AI genuinely takes work off a people team today

In the people teams I have led, repetitive administrative work took up a substantial share of the time. That is exactly where AI already delivers reliably.

Drafts. Job ads, policy documents, onboarding plans, internal communication: the first draft takes minutes rather than hours, as raw material that a person with context then sharpens. The jump in quality happens in the revision. But the blank page is gone.

Summarising and condensing. Structuring interview notes along an evaluation form, condensing several hundred free-text answers from an employee survey into themes, working the patterns out of twelve exit interviews: what used to cost days takes hours. The order matters: the AI condenses, the human interprets and decides.

Process automation, with one condition. Scheduling in recruiting, reminders during onboarding, chasing feedback deadlines: unspectacular, in total the largest time gain and the area with the lowest risk. The condition is a personal point of contact that stays. In application processes you notice very clearly when a machine is writing, and how quickly that takes the appetite out of a company. A fully automated sequence tells a candidate precisely how much interest there is in her. So every sequence needs at least one moment where a person gets in touch, however small: a short call before the first interview, a personal line with a rejection, a question in the first days of onboarding. It costs minutes and it decides how the company is remembered.

Access to knowledge. “How many holiday days can I carry over?” Questions like this tie up a surprising amount of capacity in a growing company. An assistant built on your own, well-maintained policies answers them immediately and around the clock. One condition: the sources have to be right. AI on outdated policies is just faster misinformation.

What the freed-up time is for. The real gain is not that less work gets done. It is that people work lands where it has an effect. Someone who no longer approves holiday requests and no longer types certificates has time for coaching, for supporting managers, for the conversations that otherwise get postponed. That is the difference between an administrative function and one that moves something, and it is decided by what happens to the time that was freed up. Invested in people, it pays twice: team members who are seen work better. And the people team gets back the work it signed up for in the first place. That way AI becomes an upgrade of the work itself, for everyone involved.

Where humans stay irreplaceable

The dividing line I draw in every project: AI may prepare, condense and administer. It may not assess, decide or replace where people are in vulnerable moments.

Feedback and conflict conversations. A difficult conversation works because a person holds it: with presence, with the willingness to sit through the reaction. An AI can help sort out your thoughts or prepare for the conversation. Delegating the conversation itself is cowardice with a technical finish.

Separation conversations. How a company separates says more about its culture than any set of values on a wall. I have accompanied restructurings affecting hundreds of positions. The decisive moments were always the conversations: prepared, personal, dignified. There is no use case in which a machine belongs in that room, not even in a supporting role.

Final hiring decisions. AI can structure documents and summarise interviews. The decision about whom we extend trust to and whom we bring into the team is made by a person, if only because responsibility cannot be delegated. “The system decided” is not an answer you want to give a rejected candidate or a works council. And it has a second effect: whoever makes the decision is also held to account for it. That changes the care with which it is made.

Retention. No tool binds anyone to a company. That is done by colleagues, by your own manager, by a shared purpose and by the experience of being able to move something yourself. Trimming people work for efficiency saves in exactly that place. So I ask the same question of every tool: does it leave more time for people afterwards, or less contact with them?

Use cases and their risk at a glance

Use case	Example	Risk	Human control point
Drafting text	Job ad, policy draft	Low	Review before publication
Scheduling	Interview scheduling, onboarding reminders	Low	Spot checks
Evaluating surveys	Themes from free-text answers	Medium	Interpretation and actions by the people team
Knowledge assistant	Answers to policy questions	Medium	Maintained sources, escalation to the people team
Interview summary	Notes along the evaluation form	Medium	Assessment only by the interviewer
Candidate screening	Automated ranking	High	Suggestion only, decision with a human
Performance assessment	Automated performance scores	High	Not recommended
Feedback, conflict, separation conversations	–	Very high	Human only

The path in: four steps

Most AI pilots in people work fail on missing groundwork, not on the technology. The path I work with is deliberately unspectacular:

1. Audit the repetitive tasks. For two weeks the team documents which activities recur and how much time they cost. The result nearly always surprises, and it delivers the pilot candidates for free.
2. A tightly bounded pilot. Two or three use cases from the low-risk zone, a small team, eight weeks, measurable criteria: time saved, quality, acceptance. Nothing more.
3. Guard rails and governance. Before the rollout it is put in writing: which data may go into which system? What is never automated? Who approves AI-generated content? How do we mark AI support towards team members and candidates?
4. Enablement. Only now the breadth: training, good examples, open question sessions, and with it the clear message that AI takes over routine work so that more time is left for people. Without this step every tool stays an icon nobody clicks.

A word on data protection and regulation, without wanting to be legal advice: personnel data is among the most sensitive data in a company, and the GDPR sets narrow limits on automated individual decisions. The EU AI Act classifies AI systems in the employment context (selection, promotion or dismissal, for instance) as high-risk applications with corresponding obligations. Involve data protection and, where one exists, the works council early. That is how you build the trust without which no team follows the tools.

In the end AI in people work is an amplifier: it makes good people work faster, and bad people work too. Automate the processes. Humanity is the part that needs you.