Job evaluation is the process of ranking roles by their demands — skills, effort, responsibility and working conditions — to decide which jobs count as work of equal value. Under EU Directive 2023/970, those four criteria must be objective and gender-neutral, and the grades they produce become the employer's primary defence for any pay difference between men and women.
That last clause is the change. A grading scheme used to be an internal administrative convenience. From June 2026 it is an evidentiary instrument, and almost none of the schemes now in use were built to survive being treated as one.
What the Directive actually asks of your grades
The text is short and specific. Directive (EU) 2023/970 requires that the criteria used to determine whether work is of equal value be "objective, gender-neutral criteria agreed with workers' representatives where such representatives exist," and that they "shall include skills, effort, responsibility and working conditions." It adds a sentence that is easy to skim past and hard to satisfy: "relevant soft skills shall not be undervalued."
Four further obligations hang off that definition:
- Article 5 gives applicants the right to the initial pay or pay range for the role, "based on objective, gender-neutral criteria," and bars employers from asking about current or previous pay history.
- Article 6 requires employers to make accessible the criteria used to set pay, pay levels and pay progression. Member States may exempt employers with fewer than 50 workers from the progression element.
- Article 9 sets the reporting calendar: employers with 250 or more workers report by 7 June 2027 and annually thereafter; 150–249 by 7 June 2027 and every three years; 100–149 by 7 June 2031 and every three years.
- Article 10 triggers a joint pay assessment with workers' representatives when a reported gap of at least 5% in any category of workers is not justified on objective, gender-neutral criteria and is not remedied within six months.
Read in sequence, those provisions do something unusual. They take a measurement artefact — the category of workers doing work of equal value — and make it the unit on which a statutory threshold bites. If your categories are wrong, your 5% is measuring noise.
Most member states are late, and that is not reprieve
The gap the Directive is aimed at is real and uneven. The European Commission puts the EU gender pay gap at 12% in 2023, under 5% in Luxembourg, Romania, Slovenia, Poland, Belgium and Italy, and above 17% in Hungary, Germany, Austria and Estonia — a spread wide enough that a single pan-European pay structure will meet very different local baselines.
The transposition deadline was 7 June 2026 and most of the EU missed it. A tracker maintained by Ius Laboris, updated 5 October 2026, lists only a handful of states as transposed — Italy and Slovakia among them — with drafts still moving through Bulgaria, Cyprus, the Czech Republic, France, Germany, Portugal, Spain and the Netherlands, the last of which expects plenary debate in January 2027. The Commission has not offered an extension or a simplification carve-out.
The delay is procedural, not substantive. Directive rights can take effect against public-sector employers regardless, national courts read domestic equal-pay law in light of untransposed directives, and the June 2027 reporting date has not moved. The practical reading is that employers have roughly eighteen months to make their job architecture defensible, in a period when the legal detail is still arriving.
The uncomfortable part: a grading scheme is not a measurement instrument
Here is the critique. Job evaluation has been treated for decades as a compensation-administration task rather than a measurement task, and the evidence base reflects that.
A 2025 scoping review in Compensation & Benefits Review mapped 199 job evaluation articles published between 1940 and 2023 and found that research output has declined since 2010, that methodology accounted for roughly 19% of studies, and that there is hardly any recent work on the validity or reliability of job evaluation methods — most of what exists addresses internal validity rather than whether a scheme corresponds to anything outside itself. Equal pay and gender bias made up about 15% of the literature, and the reviewers found little empirical evidence on whether these methods actually produce equal pay outcomes.
So the instrument that the Directive now relies on is one the field largely stopped studying fifteen years ago.
The second problem is that the four criteria are abstract, and abstract descriptors agree poorly. Dierdorff and Wilson's meta-analysis of job analysis reliability, published in the Journal of Applied Psychology in 2003, pooled 299 reliability estimates from 46 studies. Interrater reliability averaged .77 for task-level descriptors but only .61 for generalised work activity descriptors — the broader, more inferential kind. Rater source mattered too: incumbent ratings clustered near .77, while analyst and technical-expert ratings came out lower.
"Responsibility" and "effort" are generalised work activities. They are exactly the descriptor class where raters agree least. A scheme that scores them through a single committee conversation and no reliability check is producing numbers whose error band nobody has measured — and then asking those numbers to justify a pay difference.
Sixty-three years of precedent on the same four words
This is not a new vocabulary problem. The US Equal Pay Act of 1963 defines equal work using almost identical factors, and the EEOC's guidance is blunt about what that means in practice: jobs "need not be identical, but they must be substantially equal," and job content, not job titles, determines whether they are. Skill means what the job requires — experience, ability, education, training — not what the incumbent happens to possess. Effort is physical or mental exertion. Responsibility is degree of accountability. Working conditions covers physical surroundings and hazards.
Six decades of litigation on those words produced a consistent lesson: the fight is almost never about the formula, it is about whether the job content behind the grade was ever documented. EU employers get to inherit that lesson rather than relearn it.
Five things worth doing before June 2027
- Write down the job content, not the job title. Every grade needs a dated record of the tasks and requirements it rests on, at task level rather than at the level of a competency label. This is the single artefact that makes a grade defensible.
- Check that your raters agree. Have two independent panels evaluate the same 15–20 benchmark roles blind and compare. If agreement on responsibility and effort is materially below what you see on skills, you have found your weak factor before a regulator does.
- Treat factor weights as a declared policy choice. Weighting is not discoverable from data; it encodes what the organisation says it values. Publish the weights and the rationale. An undeclared weight is the easiest thing in the scheme to characterise as arbitrary.
- Audit for the undervaluation the Directive names. Emotional labour, service recovery, coordination across teams and attention under interruption are demands that factor plans built around physical effort and headcount tend to miss. The ILO's step-by-step guide to gender-neutral job evaluation remains the clearest practical treatment.
- Replace assumed skill requirements with measured ones where you can. If a grade claims a role needs analytical judgement at a given level, an assessed sample of that judgement is stronger evidence than a job description asserting it. AI-graded scenario assessments and dated, portable records such as AssessAll's Skill Passports produce the kind of evidence a grade can cite, rather than the kind it assumes.
One caveat against over-engineering: a 60-person employer does not need a 14-factor plan. What it needs is a written basis for its grades, the same basis applied to everyone, and someone able to explain in a sentence why two roles sit in the same band.
The takeaway
The Directive did not invent a new measurement problem; it attached consequences to an old one that most organisations had quietly left unsolved. Between now and the first reports in June 2027, the work that pays off is unglamorous — documenting job content, checking that your raters agree, and declaring your weights — because those are the three things an unjustified 5% will be judged against.