Pay transparency and AI pay analysis: opportunities and pitfalls

Short answer: The EU Pay Transparency Directive (2023/970) obliges employers to be more open about pay and must be transposed into national law by 7 June 2026. AI can make pay analysis faster and more thorough, but it can also hide inequality or introduce new bias. Anyone using AI for pay decisions must check those outcomes and comply with the GDPR.

What the directive requires

Directive 2023/970 strengthens the principle of "equal pay for equal work or work of equal value". Its core:

• Right to information: applicants learn the starting salary or salary band in advance; vacancies may no longer ask about current pay.
• Pay-gap reporting: employers above certain thresholds report periodically on the pay difference between men and women.
• Joint pay assessment: where an unexplained pay gap of 5% or more exists, the employer must, together with worker representatives, investigate and remedy the causes.
• Burden of proof: where pay discrimination is suspected, the burden of proof shifts to the employer.

Where AI helps

AI and data analysis can carry the equal-pay task. A model can cluster jobs by work of equal value, statistically adjust pay differences for legitimate factors (experience, performance) and surface the remaining, unexplained gap. That makes the reporting obligation manageable and lets an employer intervene early.

Where AI introduces bias

The same technique can also entrench inequality. If a model uses historical pay data to suggest "market-aligned" salaries, it carries existing pay gaps forward and presents them as objective. Proxy variables — part-time factor, career break, department — can correlate strongly with gender. An AI that predicts or advises on pay can therefore discriminate unintentionally, just as with AI and discrimination in recruitment. Under the directive the employer must be able to explain the difference; "the algorithm said so" is no justification.

GDPR points of attention

Pay analysis processes sensitive employee data. Points of attention:

• Legal basis and purpose limitation: salary data collected for payroll may not simply be reused for profiling or benchmarking.
• Data minimisation: use only the variables you need; avoid proxies that import discrimination.
• Transparency and rights: employees have a right to information about their pay and the criteria; see further GDPR and employee data in AI.

Practical steps for employers

• Map now which jobs are of equal value and where the pay gap sits — before the reporting obligation bites.
• Audit your pay model for disparate impact and proxy bias before use, not only when a complaint arrives.
• Keep a human in the loop for pay decisions and document the criteria applied.
• Secure the GDPR legal basis for every new analytical use of salary data.
• Involve worker representatives in good time; the joint assessment is not a formality.

Pay transparency and AI share a goal — making fair pay measurable — but only if the model is checked rather than believed.