Six of the seven pay gap metrics need more than base salary. Most employers are preparing a base salary report.

Ask an HR team in the German mid-market what data their first pay gap report needs and the answer is almost always a salary file: employee identifier, sex, function, annual gross base. That file cannot produce a compliant report under Directive (EU) 2023/970, and the shortfall is not marginal. Six of the seven metrics in Article 9(1) depend on components that are not in it.

The reason is a single definition. Article 3(1)(a) settles what "pay" means for the whole Directive, and every metric, every information right and every threshold inherits it. Get the definition wrong at the point of data extraction and the error propagates into every figure downstream, including the one that decides whether a joint pay assessment is triggered.

This article sets out what the definition actually covers, which metrics move when components are missing, where the practical data errors occur, and why the exposure is larger than a wrong percentage on a page.

The definition is one sentence, and it is deliberately wide

Article 3(1)(a) defines pay as the ordinary basic or minimum wage or salary and any other consideration, whether in cash or in kind, which a worker receives directly or indirectly in respect of their employment from their employer. The text calls everything beyond the basic wage complementary or variable components.

Three phrases in that sentence do the work. "Any other consideration" is a catch-all, not a list. "Directly or indirectly" reaches benefits routed through a third party or a group entity. "In cash or in kind" removes the distinction between a payment and a provision, which is where company cars, subsidised meals, and housing allowances enter the scope.

Recital 21 then names examples of complementary and variable components: bonuses, overtime compensation, travel facilities, housing and food allowances, compensation for attending training courses, payments in the case of dismissal, statutory sick pay, statutory required compensation, and occupational pensions. The recital says these components may be included among others, so the enumeration is illustrative rather than closed. Anything of value flowing to the worker because of the employment relationship is in scope until you can articulate why it is not.

Two items on that list surprise most practitioners. Overtime compensation is pay, which means a workforce where men systematically book more paid overtime carries a structural gap that a base salary report cannot see. Occupational pension contributions are pay, which means an employer contribution that varies by grade, by tenure, or by contract type feeds directly into the reported figures.

The definition contains no separate rule for part-time workers, and it does not need one. Article 3(1)(b) defines the pay level as gross annual pay and the corresponding gross hourly pay. The hourly figure is what makes a workforce with uneven working-time patterns comparable, and it is derived from the same total, not from base salary.

Which of the seven metrics move, and how

The seven metrics of Article 9(1) are not seven views of the same number. They sit at different levels and use different pay bases, and a base-only extract fails them in three different ways.

Point (f) deserves a second look, because it fails quietly. Quartiles are formed by ranking every employee by pay level and cutting the workforce into four equal bands. If the ranking uses base salary while the Directive requires total pay, employees with large variable components move bands. The reported proportions then describe a distribution that does not exist, and nothing in the output signals a problem.

Point (g) is the one that carries legal consequence. It is the only per-category metric, and it is the metric the Article 10 trigger reads. A joint pay assessment becomes due only when three conditions hold together: the average pay level of female and male workers differs by at least 5% in any category of workers, that difference is not justified on objective, gender-neutral criteria, and it has not been remedied within six months of the report. A category whose gap sits near 5% on base salary alone can cross the line or fall below it once bonuses and pension contributions are included.

Where the data actually goes wrong

The definitional question is settled by reading the text. The practical failures happen further downstream, in the extract itself, and they cluster in four places.

The second failure is scope drift between systems. Base salary lives in payroll, bonuses often live in a separate incentive tool, pension contributions live with the provider, and benefits in kind live in a spreadsheet on someone's desktop. Each source has its own reference period and its own population. A component that exists but is not extracted is indistinguishable, in the final report, from a component that does not exist.

The third is the treatment of employees who did not receive a component. For metric (e) the distinction between a zero and a blank is the entire measurement: the metric asks what proportion of each sex receives variable components. An employee with no bonus is a zero, an employee whose bonus was not extracted is a blank, and treating the second as the first inflates the denominator of a metric that is meant to expose exactly that asymmetry.

The fourth is normalisation. Pay is reported as gross annual and gross hourly. Monthly figures, part-time salaries, and mid-year joiners all need converting before comparison, and the conversion has to be applied to every component consistently. A bonus that was annual and a base that was monthly cannot be added without a decision, and the decision belongs in the methodology note rather than in a formula nobody can find.

Why the scope decision is a documentation problem

A wrong percentage is recoverable. You correct it, you explain it, you move on. The more durable exposure is that the Directive gives other people the right to inspect how the number was made.

Article 9(6) requires the employer's management to confirm the accuracy of the information after consulting the workers' representatives, and gives those representatives access to the methodologies applied. Article 9(10) gives workers and their representatives the right to ask for clarifications and details on the reported data, including explanations of any gender pay differences, and obliges the employer to reply within a reasonable time by means of a substantiated response.

That means the list of components you included, and the components you excluded, becomes a document that is read by the works council rather than an internal choice. A report which states that occupational pensions are out of scope, while the underlying calculation quietly includes an employer pension contribution, is a false statement about method in exactly the document whose method is open to inspection.

Article 18(2) is what gives that its weight. Where an employer has not fulfilled the transparency obligations set out in Articles 5, 6, 7, 9, and 10, the burden of proof shifts to the employer in equal pay proceedings without the worker having to establish a prima facie case first, unless the breach was manifestly unintentional and minor. Poor documentation is therefore not a hygiene issue that sits beside the legal risk. It is one of the routes into it.

What to settle before you extract payroll data

The order matters, because the scope decision constrains the extract, and the extract constrains everything after it.

1. Write the component inventory first. List every element of remuneration your organisation pays, in cash or in kind, and mark each one in scope or out with a one-line reason. Do this before anyone opens the payroll system.
2. Assign each in-scope component to base or complementary. This is the split that point (g) requires. Ordinary basic salary on one side, everything else on the other.
3. Find the components that do not live in payroll. Incentive plans, pension contributions, company cars, allowances, training compensation. These are the ones that are missing rather than wrong, and they are missing silently.
4. Check the unit of every column before mapping it. Currency or hours, gross or net, monthly or annual, full-time equivalent or actual. Overtime is where this fails most often.
5. Decide how a non-recipient is represented. A true zero and an absent value are different facts, and metric (e) reports the difference between them.
6. Normalise once, centrally. Gross annual and gross hourly, per Article 3(1)(b), applied to every component by the same rule and documented in one place.
7. Put the inventory in the methodology note. The works council will see it under Article 9(6) either way. A note that arrives with the figures is a different conversation from one produced under a request for clarification.

None of these steps requires a legal opinion. They require somebody to write down what the organisation pays and where each element is stored, which in most companies of 250 to 2,000 employees has never been done in one document.

The Directive's definition of pay is not an edge case buried in the definitions article. It is the input to every figure that will be reported, published, compared across employers by the monitoring body under Article 29, and, for point (g), given to every employee under Article 9(9). An organisation that has not settled what counts as pay has not started the calculation, whatever its salary file says.