How the figures are produced

The data layer

The same governed people layer the platform’s HR & People module reads — the one that carries headcount and FTE, utilisation, absence and turnover. Pay transparency does not get a layer of its own; it gets this one, taken to the depth the act requires.

Every pay component from payroll, HR and the year-end annexes, mapped into one model held in your own subscription. One row per employee per component, which is what makes the aggregation honest: components are summed per employee before any comparison, so nobody is counted twice because they happen to have four pay lines and their colleague has two.

Inputs are versioned. Any cycle can be reproduced from the data that produced it, which is what makes a report from two years ago defensible when someone asks about it.

The pay-component taxonomy

Base pay, variable pay, thirteenth and fourteenth salary, allowances, overtime and jubilee payments, and hourly rates — each classified into a pay group and each split between target and paid.

The taxonomy matters more than it sounds. The act reports base pay and complementary pay separately — so a component sitting in the wrong group moves both figures at once.

The reported figures

Gaps are expressed as the difference between women and men relative to men, so a negative figure means women are paid less.

Explaining the gap

A raw gap says how much. It does not say why — and the act asks employers to justify differences on objective criteria.

The model runs a Blinder-Oaxaca decomposition with grade and tenure as regressors, and reports four things: how much of the gap those factors account for, how much they do not, the absolute amount of the unexplained part, and the fit of the underlying model so you can judge how much weight the split carries.

The unexplained residual is the part that has to be justified, and the part worth spending remediation budget on. The explained part usually points somewhere else — at how people were graded, or at who was hired into which role.

Alongside the raw figures the model reports an adjusted set: what each gap would be if women and men had the same grade and tenure profile. Read against the raw number it separates two questions that get argued as one — whether people doing comparable work are paid comparably, and whether the two groups are distributed across the structure the same way. Those have different answers and different remedies.

The mean and the median are a diagnostic, not a duplicate. The act requires both, and the distance between them is informative on its own. When the average gap is materially wider than the median, a small number of high earners is pulling it — which is a question about who reaches the senior grades, not about how people are paid within them. When the two sit close together, the gap runs through the whole distribution and the pay-setting itself is where to look.

Finding the cause

A decomposition says how much of a gap is unexplained. It does not say where it came from, and that is the question you have to answer to either justify a difference or fix it.

The analysis works down four levels:

• By category, ranked by impact. Categories are ordered by how much each one moves the company figure — gap size against headcount — not by the size of the percentage. A wide gap over eleven people and a narrow one over two hundred are not the same problem, and sorting by percentage puts them in the wrong order.
• Within grade. A gap that survives when grade is held constant is a pay-setting difference between people doing comparable work at the same level. That is the finding with the least room to explain it away, and the one to look at first.
• At the entry point. Where people arrive on the scale at hire, because percentage raises compound whatever difference is there on day one. A gap created at hiring is invisible in this year’s decisions and permanent in the numbers.
• Individually. Pairs at the same grade, in the same category, in the same part of the business, with a difference that nothing in the data accounts for. These are rare, they distort the category figure out of proportion to their number, and each one needs a written justification or a correction.

What the analysis rules out matters as much as what it finds. Where grade, tenure, hours or the benefit structure turn out not to account for a gap, that is recorded too. When someone asks why a difference exists, being able to show what was tested and eliminated is a large part of the answer — and it is the part nobody has when the report was assembled by hand.

Costing the remediation

Two strategies, computed per category that breaches, reported side by side.

Uniform uplift. The percentage raise the underpaid gender’s median needs to come within threshold, applied to everyone of that gender in the category. It closes the gap by construction and it is easy to communicate. It also pays people who did not need it.

Floor adjustment. A floor set at the other gender’s median less the threshold, with only the individuals below it lifted to it. It is targeted and it costs less per person — but it does not guarantee the median moves far enough to clear the threshold.

Each is reported with how many people it actually touches and what the average adjustment is, because that is a different decision from the total. One is a policy you announce to a whole group; the other is a set of individual conversations. They can cost about the same and be nothing alike to carry out.

Which one is cheaper depends on how wide the pay distribution is, and it changes category by category. A wide distribution with a long low tail can make the targeted strategy the more expensive one, which is counter-intuitive and worth seeing before a budget is set.

The scenario modeller runs both, plus reclassification of roles that were graded wrongly — which carries no salary cost at all — with the breach count and the total updating as you move.

Reclassification is proposed only where the analysis finds roles sitting outside the normal spread of their category, and it is only accepted if it does two things at once: reduce the total number of categories over the threshold, and leave the receiving category still under it. Moving a problem from one category into another is not a fix, and a tool that lets you do it by accident is worse than no tool.

The plan that comes out of it

The figures are the obligation. What you do about them is the work, and the model produces that as an output rather than leaving it as a reading exercise.

Each finding comes out with a severity, a timeframe and what it is for: individual cases that need a justification written or a correction made now; categories that need a joint pay assessment opened; policy questions — how variable pay is awarded, where people enter the scale — that take a cycle to change and cannot be fixed with a payment; and the structural work on progression, which is slower than everything else and is what stops the gap coming back.

The remediation budget phases against that order rather than arriving as one number. Sequencing matters as much as the total: the severe and the individual cases first because they carry the most risk per euro, the policy changes next because they cost little and prevent recurrence, and the structural spend last because it is the largest and the least urgent. A single total tells you what it costs. The phasing tells you what to do in the first quarter.

The point of pricing remediation is not the number. It is that the number can be put in a budget, in an order somebody can defend.

The disclosure rule

The act sets a test about identifiability, not a minimum group size. The threshold applied is therefore a judgement, and it is recorded as a documented parameter with the reasoning behind it.

Every figure that fails the test is suppressed, in the ministerial report and in the answer to an employee request alike. A median computed over a handful of people discloses their pay to each other — and the same rule has to apply in both places, or the two disagree.

A second and separate floor sits underneath it: a gap between one woman and one man is arithmetic, not a statistic, and the model will not report it as one. Where either group is too small for the comparison to mean anything, the figure is not computed rather than computed and shown with a caveat. The two rules answer different questions — one is about protecting an individual, the other about whether the number says anything — and conflating them produces figures that are either unsafe or misleading.

The individual view

One page per employee: their pay against the median, quartile and range of their peer group, at grade, cluster or category level, with a compa-ratio and a percentile rank. Peers can be read at any of the three levels, because the right comparison depends on what is being asked.

It reads the same model that produces the ministerial report, so an employee’s answer and the filed figure cannot drift apart.

Where the boundary is

Legal interpretation rests with your counsel. We produce data and reports; we do not advise on the act and we do not certify compliance.

The job evaluation stays with the employer. § 3(1) names complexity, responsibility, effort and working conditions, and adds soft skills. We facilitate the exercise, document it and structure the model around the result, with a compensation-methodology specialist alongside where that support is wanted. The classification itself is a decision the employer defends.

The pay decisions — remediation, the justification for any difference, and the conversation with employee representatives — are yours.