What can employers learn from UWV's scrapped matching project?

Today UWV matches jobseekers to vacancies mostly on education and work history. The new service was supposed to add skills, competences and ambitions, so that someone would also be shown a vacancy in a different occupation or sector. It is a sound idea, and UWV says the ambition stays.

The trouble was in the delivery. On 4 February 2026 the Dutch government's ICT review board found the project was not under control. Its advice was strikingly practical: first show that the matching itself adds value, and only then keep building. UWV went on to test the service on a small scale. An external quick scan then found, in the minister's words, significant bottlenecks in four areas: data quality, explainability, architecture choices and governance. UWV itself points to the data quality it needed and the technical fit with its existing environment.

Not everything is lost. Part of the money went into an integration platform and a combined database of vacancies and profiles, and according to UWV a considerable part may be reusable. The service it was all for, though, will not arrive in this form.

Because matching on skills needs skills that are recorded somewhere. Most job adverts contain a title, a level of education and a list of tasks. Most candidate profiles are a CV in free text. Software that has to match on skills must first work those skills out for itself, and it does that only as well as the text it reads.

Eurostat's figures show this is not just a UWV problem. Of businesses in the Netherlands with ten or more employees that considered AI but do not use it, 36.1 per cent cited problems with the availability or quality of the data needed in 2025. Another 33.7 per cent cited incompatibility with existing equipment, software or systems. Those are the same two reasons UWV gives, only on a larger scale.

The lesson from UWV is not to stay away. It is to reverse the order: prove first, build second. That comes down to five things.

Prove that the matching adds something. Take a vacancy you fill often and put the software's shortlist next to your recruiter's. Agree in advance what you count: how many candidates reach an interview and how many are hired. It is how we measure the results of Thought Leader Recruitment too: every step traceable, from first contact to hire.

Start with your own data. Are the skills you are looking for written into your vacancies, or only job titles? Do you know what your current people can do? Without that base, the software matches on guesses.

Sort out the integration before you buy. Ask how the software reads your applicant tracking system and what comes back. A standalone tool your recruiters have to feed by hand will be abandoned within three months.

Insist that you can explain why someone ranks first. If the supplier cannot say why candidate A sits above candidate B, you cannot explain it to a rejected applicant either. That already touches the GDPR, and from December 2027 the European rules on AI in hiring as well. What applies when is set out in Which EU rules apply to AI in hiring right now? And now and then, read a sample of rejected candidates by hand; that is the weak spot of every system that filters candidates.

Decide in advance when you stop. UWV stopped after five years. Before you start, decide what outcome you want to see after a trial period, and what you will do if it is not there.

Is matching on skills a bad idea, then?

No. UWV is dropping this system, not the idea: according to UWV, the ambition to put skills and competences first remains. Looking at what someone can do rather than at their last job title often turns up candidates an ordinary search misses. The point is that the software doing it is only as good as the data behind it and the people who work with what it produces.

Should I buy this kind of software myself?

Not necessarily. For an employer with a handful of vacancies a year, an in-house system with integrations, maintenance and checks rarely pays for itself. If you hire on a structural basis, you can also leave the matching to a partner that runs the process under your name. How that works is explained in what is RPO. Ask that partner the same questions, though: which software has a say, and can anyone explain how?

Does matching software replace a personal approach?

Not for scarce profiles. Software finds people who have already put themselves forward as candidates, or whose profile looks similar. The people you need most are often registered nowhere and only respond to someone they know or trust. How to build that group step by step is shown in how Thought Leader Recruitment works, and how it plays out per situation in the Thought Leader Recruitment cases.

SocialFind recruits for employers who need people on a structural basis, and wrote this article from UWV's own announcement, the letter to the Dutch Parliament of 14 October 2026 and Eurostat's figures, checked on 29 October 2026. If you would like articles like this in your inbox, sign up for our newsletter.