Compensation Benchmarking Agent

Pay transparency is arriving with a hard deadline

This agent follows the Decision Layer principle: each decision is either rule-based, AI-assisted, or explicitly assigned to a human.

By June 2026, organisations with 100 or more employees must submit gender-specific pay reports for the first time under the EU Pay Transparency Directive (2023/970). Where a report shows an unexplained gap of 5 percent or more within an employee group, the employer has six months to remediate - or must initiate a joint pay assessment with employee representatives. The burden of proof reverses: employees no longer need to demonstrate discrimination; the employer must demonstrate its absence.

That is the regulatory framework. The operational reality looks different.

Compensation data without a compensation system

16 percent unadjusted gender pay gap is the figure that Eurostat reports across the EU. Adjusted - meaning for comparable role, qualification, and employment history - the gap narrows but does not vanish. For many organisations, however, the real challenge is not the headline statistic but the granularity: how does the gap distribute across 80, 200, or 500 different roles? Which positions are below market? Where does turnover risk arise because compensation is no longer competitive?

The answer requires compensation surveys. Mercer, Willis Towers Watson, Radford, and others deliver annually updated market data. The bottleneck is not the data itself but the matching: which internal role corresponds to which market benchmark? Job titles are company-specific. Responsibility levels rarely align one-to-one. The same position carries different names across three survey providers.

In most Comp&Ben departments, the work looks like this: once a year the survey arrives. Someone opens a spreadsheet. Begins matching role by role. Calculates compa-ratios - the ratio of what an employee earns to the market median for their role. Builds pivot tables. Finds inconsistencies. Corrects mappings from the prior year. Loses three weeks. The result is a snapshot that is already outdated when completed.

The Pay Transparency Directive demands continuous data, but spreadsheets deliver only annual snapshots

The Pay Transparency Directive does not ask for just any figures. It demands average compensation per comparison group, broken down by gender, verifiable for every individual employee who requests disclosure. Organisations with 250 or more employees report annually. Between 100 and 249, every three years.

This means: the compensation structure must be queryable at all times. Not once a year when the survey arrives. Not when the board asks. Always. Organisations that still compute compa-ratios in manual spreadsheet rounds face a capacity problem that overtime cannot solve.

There is a second requirement. The directive mandates that candidates must be informed about the salary range for the advertised position before the first interview. This presupposes that such ranges exist, are current, and are grounded in verifiable market data. (US: Similar requirements already apply in several states, including Colorado, New York, and California.)

From annual project to continuous process

A Compensation Benchmarking Agent changes the operating logic. Instead of an annual analysis, a continuous process emerges where every step follows a clear decision principle.

Market data          Job matching         Compa-ratio          Deviations
import          -->  AI assigns,     -->  calculated      -->  prioritised
(human selects       human validates      automatically        by risk
 sources)            critical cases       (rules engine)       (AI + human)

In the first step, Comp&Ben decides which external sources are relevant - which surveys, which markets, which comparison groups. No automation takes over this selection. In the second step, the agent matches internal positions against external benchmarks on the basis of job descriptions, responsibility level, and job family. Critical mappings - leadership positions, specialist roles, new positions without a market equivalent - go to manual review. Compa-ratios are calculated rules-based, no AI required: actual pay divided by market median, per position, per location, per gender.

The decisive difference lies in the fourth step. Positions with significant deviations are not buried in a spreadsheet but reported with priority: by turnover risk, by proximity to the directive’s 5-percent threshold, by the number of affected employees. The analysis delivers recommendations but makes no compensation decisions. Whether bands are adjusted is a human decision.

Pay transparency as an organisational challenge

The directive does not only affect Comp&Ben. It touches recruiting (salary ranges in job postings), HR controlling (reporting obligations), employment law (burden of proof reversal), and employee representatives (joint pay assessment where anomalies surface). Treating this as an isolated compliance project builds silos rather than structures.

Compensation bands maintained by the benchmarking agent are the foundation for every subsequent pay process: promotion decisions, merit cycles, salary negotiations in recruiting. The data structure that emerges here - job families, levelling, market reference points - becomes the backbone of the entire compensation architecture.

The question is not whether organisations must make their compensation data transparent. The deadline is set. The question is whether the data will be defensible when it is.