Hiring · August 1, 2026 · 9 min read

Hiring in France: skills-based assessment in a talent drought

Hiring in France means facing a skills drought and EU rules at once. Skills-based assessment widens a thin pool while staying defensible under the EU AI Act.

← Part of The five pillars of hiring: what assessments measure

This guide is for the hiring managers, talent leaders, and recruiters staffing roles in France in 2026, and for anyone outside the country trying to hire French-based talent into a European team. The stakes are unusually high right now because two problems have arrived at once. The first is global and you already feel it everywhere: the signals a CV used to send have decayed, so the degree, the school, and the tidy career arc no longer tell you who can actually do the work. The second is local and sharper. France is in a real skills drought, and the old pedigree filter does not just misread candidates any more; it actively shrinks a pool that is already too thin to fill your open roles. The way out of both problems is the same one: assess for demonstrated skill, widen who you can consider, and do it in a way that stays defensible under tightening EU rules. That is what skills-based hiring in France now demands.

Why is hiring in France so difficult right now?

Because the shortage is measurable and the usual filter makes it worse. As of 2026, 72% of French companies cite skilled-worker shortages as a barrier to investment. There are more than 15,000 unfilled cybersecurity roles, and STEM enrolments sit below the EU average, so the pipeline feeding those roles is not refilling fast enough. Zoom out and the picture is continental: Optima Europe reports more than 500,000 unfilled tech jobs across Europe in 2026, with 57% of EU firms unable to find qualified technical staff. France is not an outlier here; it is a well-documented example of a Europe-wide squeeze.

Here is the trap. When talent is scarce, the instinct is to raise the bar on credentials, to filter harder on the grande ecole, the right degree, the recognisable employer. That instinct backfires. Every credential you require removes people who can do the job but never had access to that credential, and in a drought you cannot afford to remove them. The scarcer the talent, the more expensive it is to keep screening for pedigree instead of ability. The general problem of CV noise and the local problem of a thin pipeline point to the same fix, which is the subject of the skills-based hiring guide.

Why have CV signals stopped working?

The CV was always a proxy. A degree stood in for knowledge, an employer's name stood in for a hiring bar someone else set, and years of tenure stood in for competence. Those proxies were leaky before; now they are close to broken. Applications are generated at scale, keywords are optimised for whatever an automated filter is thought to want, and a polished document tells you almost nothing about whether the person behind it can debug a real system or handle a real customer. Spotting a machine-written application has quietly become part of the recruiter's job, which is itself a sign of how far the old signal has decayed. In a drought, spending your best hours forensically reading CVs is doubly wasteful: the document is less trustworthy than ever, and the time spent squinting at it is time not spent assessing the people who could actually fill the role.

The honest response is not to hunt harder for the perfect proxy. It is to stop relying on proxies and measure the thing itself. If you want to know whether someone can do the work, give them a small, realistic slice of the work and watch what happens. That is the core of work sample tests, and it is the single most predictive move available to you. It also happens to be the move that widens your pool rather than narrowing it, which is exactly what a drought-constrained market needs.

How does skills-based assessment widen a drought-constrained pool?

It works by changing what you filter on. When the first gate is a standardised, job-relevant assessment that every candidate takes on the same terms, the question stops being 'where did you study' and becomes 'can you do this task'. That reframing quietly lets in the people a credential filter locks out: the self-taught developer, the career changer moving in from an adjacent field, the person with a strong portfolio and an unremarkable degree. In a country with 15,000-plus unfilled cybersecurity roles, those are not fringe candidates. They are a large share of the people who can actually fill the gap.

Skills-first hiring also compounds. A fairer front gate produces a more diverse funnel, which over time builds a workforce that does not all look and think alike. It is worth reading reduce bias in hiring alongside this, because the mechanism that widens your pool is the same one that reduces the risk of screening out protected groups. Widening and fairness are not competing goals here. They are two effects of the same decision to measure ability directly.

There is a practical objection worth answering. Managers in a hurry often say they do not have time to run assessments when roles have been open for months and the pressure to fill them is intense. But the time maths runs the other way. A thin pool filtered on pedigree produces a trickle of look-alike candidates and a long, anxious search; a wider pool filtered on demonstrated skill produces more viable people, faster, and a shorter path to a confident yes. In a market where 72% of French firms name the shortage as a barrier to investment, the assessment is not the thing slowing you down. The credential filter is.

What does the EU AI Act mean for hiring in France?

This is where the second half of the France problem lives. You cannot solve the drought by reaching for any assessment tool you like, because employment AI is classified as high-risk under the EU AI Act, and France applies the Act like every member state. The obligations that matter to employers using AI in candidate evaluation are documentation, human oversight of decisions, and transparency to candidates. Get the assessment approach right and defensibility is a natural by-product; bolt AI onto a fuzzy process and you inherit risk you did not have to take on.

The dates matter, so use them precisely. Under the Digital Omnibus deferral, in force from 27 July 2026, the high-risk employment-AI obligations bite from 2 December 2027. Separately, Article 50 transparency duties apply from 2 August 2026. Our EU AI Act hiring post is the place to align the detail; treat these dates as the version of record for France as well as the wider EU, and design your process to meet them rather than scrambling to retrofit later.

How do you assess French language for a specific role?

If you are hiring foreign talent into a French-based role, the language question deserves care, and 2026 added context to it: from 1 January 2026, France requires B2 French, written and spoken, for naturalisation. That is a citizenship threshold, not an employment one, and conflating the two is a common and costly mistake. What a person needs to become French has little to do with what a particular job requires day to day, so treat the naturalisation rule as background, not as a hiring bar.

For the job itself, assess the work language a person will actually use. A DELF or DALF certificate is a useful data point, but a certificate measures general proficiency at a moment in time, not whether someone can run a standup, write a clear incident report, or handle a frustrated client in the language your team works in. Read DELF and DALF for hiring for how those levels map, and CEFR levels for hiring for the framework underneath them, then design a short, role-shaped language check for the tasks that matter. Certificates set a floor; the work sample tells you the truth.

How do you keep AI-based assessment defensible in France?

The reassuring news is that the thing which makes an assessment predictive is largely the same thing that makes it defensible. A job-relevant, standardised evaluation that every candidate takes on equal terms is both a better signal and a stronger legal footing than an unstructured chat, because it measures the role rather than the resemblance to the interviewer. Build the process around demonstrated skill and you are already most of the way to meeting the EU AI Act's demands for fairness and consistency.

• Keep a human in the decision. AI can generate and help evaluate a job-relevant assessment, but a person should own the hire or no-hire call, which is what the high-risk oversight expectations point toward.
• Standardise the assessment so every candidate faces the same task on the same terms, giving you both a cleaner signal and a fairer, more auditable process.
• Document what you assess and why. Being able to explain the logic of an evaluation is central to the transparency direction of the Act.
• Audit for adverse impact. Widening the pool is only a win if you check that the assessment is not quietly screening out a protected group; the four-fifths rule is a practical starting test, and pass rates by group are worth watching over time.
• Disclose the use of AI to candidates where required, in line with Article 50 transparency from 2 August 2026, and keep the records that let you show your working.

None of this is exotic. It is the discipline of an AI-native skills assessment platform used well: AI-generated assessments that are standardised, documented, human-supervised, and audited. Done properly, defensibility stops being a compliance tax and becomes a side effect of running a rigorous evaluation. If you want to see the shape of it, book a demo or run through a sample assessment yourself before you put a candidate through one.

Where does this leave a hiring manager in France?

In a better position than the headlines suggest. The drought is real and the rules are tightening, but the response to both is a single, coherent move: stop filtering on proxies and start measuring skill. That decision widens a pool that a credential filter would strangle, it produces a fairer and more diverse funnel, and, built with a human in the loop and a documented, standardised process, it holds up under the EU AI Act rather than fighting it. The French market rewards employers who treat assessment as their core capability, not an afterthought.

The employers who win in France's talent drought are not the ones guarding the highest credential bar. They are the ones who stopped asking where a candidate came from and started measuring what a candidate can do, then built that measurement to be fair, documented, and defensible from day one. Scarcity does not reward gatekeeping. It rewards a better way to see skill.