Klikwork
Opinion

Recruitment has no R&D department

·10 min read
Marcel van der Meer
Marcel van der MeerFounder, Klikwork
Five lit workbenches in a dark room and an empty orange outline where the sixth should stand

TL;DR

HR is last on AI adoption in Dutch organisations, at under 10 percent while every function with a sanctioned place to experiment sits between 22 and 30 percent. That is not a skills gap. Recruitment has no hack week, no test log, no pilot slot, so AI enters individual people instead of the department. Fix the missing room, not the missing tool.

Every other function has a place where things get tried. Talent acquisition does not, and that is most of the reason your team is uneven.

Ask an engineering team where new ideas get tested and they will point at something. A hack week. A spike in the backlog. A Friday that is protected for it. Ask marketing and they will show you a test log. Ask sales and there is a pilot running with two reps before anything reaches the floor.

Ask a talent acquisition team the same question and you get a pause.

There is no place. No budget line, no slot in the calendar, no permission. So when something like AI arrives, it does not enter the department. It enters individual people, one at a time, in the gaps between their vacancies.

That is a structural fact, not a character flaw, and it explains almost everything about how AI has landed in recruitment.

HR is last, and it is not close

Berenschot and Waag Futurelab asked Dutch organisations which departments actually use AI. The 2024 ranking:

  • Communication, customer service and support: 30 percent
  • Research, development and innovation: 29 percent
  • IT and technical support: 27 percent
  • Sales and marketing: 22 percent
  • HR: under 10 percent

Every function that has somewhere to try things is at the top. The one function that has nowhere is at the bottom, at a third of the rate of the next-lowest.

A separate survey by Visma YouServe among more than 500 HR professionals found that 35 percent of HR departments use no AI at all.

Meanwhile the organisations around them are moving. CBS measured AI use among companies with ten or more employees at 22.7 percent in 2024, up from 14 percent the year before. In the 50 to 250 band it went from 20 percent in 2023 to 45 percent in 2025. Among companies with 500 or more employees it was already 59.2 percent in 2024.

So the building is adopting AI. The talent team inside it is not.

Encouragement is free. Room is not.

UWV surveyed 3,551 employers in the autumn of 2025, representative for Dutch employers across nearly all sectors. The share using AI to a reasonable or high degree doubled in a year, from 16 to 32 percent. The share using none at all halved, from 60 to 34 percent.

Six in ten of the employers who use AI offer their people some form of guidance. Which leaves four in ten offering none.

The interesting part is what the six in ten actually do. In UWV's own words: employers mainly let employees learn from each other and encourage them to experiment by themselves, and to a lesser degree set guidelines or offer training and courses.

Now hold that against what employees report. Newcom surveyed 3,122 Dutch adults in July 2026. 5.2 million people use AI at work and almost six in ten of them have it in their daily routine. Only three in ten ever received training or instruction. Only four in ten say their organisation has any rules about AI use at all.

Encouraging experimentation costs nothing, which is why it is the most popular form of support. Making room for it costs time, and time is the one thing a recruitment team never has spare. So the encouragement is real and the room is not, and everybody quietly understands the difference.

Berenschot found the end state of that: at 9 percent of organisations was AI used structurally, on the basis of an actual AI policy. More than half had no guidelines for employees at all.

What that produces inside a team

Two or three people do it anyway. On their own account, in their own evening, on a task they chose themselves. They get good at it, quickly.

The rest wait. Not out of resistance. They are busy, nobody gave them an hour, and nobody told them where to start. Six months in, the gap is wide enough that the ones who are ahead stop asking the others and start routing work around them.

At that point the department has not adopted AI. It has split.

And the split costs more than the delay did. Two speeds in one team means two standards, two sets of assumptions about what is allowed with candidate data, and a manager who reads silence as agreement.

Why the three usual fixes miss

Send one person on a course. They come back fluent, and the distance to everyone else is now larger than before they left. You have improved an individual and worsened a team.

Roll out a tool. A licence is not a practice. Most people log in once, do not know what to ask it, and never open it again. The ones who already experiment were not waiting for your licence.

Write a policy. Necessary, and more than half of organisations still do not have one. But a policy tells people what they may not do. It has never once made anybody try something.

All three treat the problem as knowledge or access. The problem is that nobody was ever given a sanctioned hour in which failing was fine.

What an R&D function looks like when nobody will fund one

You are not going to get a lab. You can get five things, and none of them cost much.

  1. A slot that is not a meeting. Blocked, in the calendar, with a start and an end. Not "when it is quieter", because it is never quieter.
  2. A real task, not a demo. The one thing on their list they do every week and dislike every week. Demos teach people what a tool does in somebody else's hands. Their own annoyance teaches them what it does in theirs.
  3. More than one person. One person experimenting produces a private habit. Two or three produce something the rest of the team can copy, and somebody to ask when it breaks.
  4. A deadline. Unbounded exploration ends nowhere. A finish line, even an artificial one, is what turns interest into a result.
  5. Someone who decides afterwards what survives. This is the step almost everyone skips. Without it, everything built in the sanctioned hour quietly dies during the first busy week, and the conclusion becomes "we tried that, it did not stick".

That is the whole design. It is not sophisticated and it does not need budget approval. It needs one manager willing to say that for these hours, output is allowed to drop.

The part I will not pretend about

This does not work if you do it once. A single afternoon produces a pleasant afternoon.

It also does not work if you only invite the enthusiasts. They will have a good time and the gap will widen, which is the exact outcome you were trying to avoid.

And it will not survive a locked-down environment unless somebody checks the access beforehand. There is nothing more deflating than a room full of people who cannot log in.

The point

Recruitment is one of the last functions in the business without a place to try things, at the exact moment when the ability to try things is the whole game. Another tool will not fix that, and neither will another policy.

It gets fixed by putting the entire team in a room, on their own work, with a deadline, and deciding afterwards what stays.

That is what Hack your recruitment week is: two days in your own office, your whole talent team, on the tasks in your week you dread most.

Frequently asked questions

Which departments use AI the most, and which the least? In Dutch organisations the ranking for 2024 was communication, customer service and support at 30 percent, research, development and innovation at 29 percent, IT and technical support at 27 percent, and sales and marketing at 22 percent. HR came last at under 10 percent. Source: Berenschot and Waag Futurelab, AI-Trendonderzoek 2025.

How many Dutch companies use AI? 22.7 percent of companies with ten or more employees used at least one AI technology in 2024, up from 14 percent in 2023. Among companies with 50 to 250 employees, use rose from 20 percent in 2023 to 45 percent in 2025. Among companies with 500 or more employees it was 59.2 percent in 2024. Source: CBS.

Why is HR behind on AI compared to other departments? Not because HR people are less capable or more cautious. Every function that ranks above HR has a sanctioned place to experiment: engineering has hack weeks, marketing has tests, sales has pilots. Recruitment has no equivalent, so AI never enters the department as a shared practice. It enters individual people privately.

Do employers actually train their people on AI? Mostly not. Six in ten employers who use AI offer some form of guidance, but that guidance mainly consists of letting employees learn from each other and encouraging them to experiment on their own. Setting guidelines and offering training happen less often. On the employee side, only three in ten working AI users have ever had training or instruction. Sources: UWV employer survey 2025, Newcom AI Monitor 2026.

How many organisations have an AI policy? At 9 percent of organisations AI was used structurally on the basis of an actual AI policy. More than half had no guidelines for employees at all. Only four in ten employees say their organisation has any rules about AI use. Sources: Berenschot and Waag Futurelab 2025, Newcom 2026.

Is sending one recruiter on an AI course enough? It usually makes the team less even, not more. The person comes back fluent and the distance to their colleagues is larger than before they left. If the goal is a department that works to one standard, the unit of training has to be the team, not the individual. That is the reasoning behind Hack your recruitment week.

What is an AI hackathon for a recruitment team? A time-boxed session, normally one or two days, in which an entire talent team builds something on their own vacancies and their own workflows instead of listening to a presentation. The point is not that everyone becomes a builder. The point is that everyone has built something once, together, so the practice is shared rather than private. Ours runs as Hack your recruitment week, two days with two trainers, in your own office.

How long does it take before a talent team is actually using AI? Adoption is not really a function of time, it is a function of permission. Teams that are given a protected slot, a real task and a deadline move within days. Teams that are only encouraged to experiment in their own time can stay uneven for years, which is what the adoption figures above show.

Does this only work if we have modern tools? No, and it should not. Most corporate talent teams work inside restrictions their IT department set, often with Microsoft Copilot and little else. The exercises have to be designed so the thinking survives the tool. What does matter is checking beforehand what the team can actually log in to.

Sources