The weighing says A, you agree on B. Who establishes the classification?

You have weighed your roles. The result puts two of them in different categories. And around the table, with your works council present, everyone agrees that in practice those two roles are of equal value.

So who is right?

This question comes up in almost every job architecture project, and it is less technical than it looks. The answer decides what your instrument actually is: a decision maker or an aid.

A classification is a decision, not an output

The Dutch pay transparency bill (Kamerstuk 36949, implementing Directive 2023/970) asks the employer to establish categories of workers performing equal or equivalent work. That verb is the whole point: establish. It does not say calculate.

That is how we read the explanatory memorandum, at least, and it matches where the works council enters the picture. Over the job evaluation system itself the works council has no right of consent as long as the system is settled in the collective agreement, because that is where the unions sit. But in establishing the categories, the objective pay criteria and the approach to unjustified pay differences, the works council is involved. The layer your collective agreement does not deliver is precisely the layer your works council has a say in. That is worked out further in we are covered by a collective agreement, does that settle pay transparency.

Something practical follows from that. If the employer and the works council jointly establish that two roles belong in the same category, then that is the classification. A scoring model that rides over it does not produce a better measurement. It produces an answer nobody in the room recognises.

Why the order matters

Here is the mistake that is easy to make, and we made it ourselves first.

The intuitive way to build this is: let the model group every role by its weighing, then give the user a button to adjust the result afterwards. That sounds reasonable and it does not work. The model groups by relative distance. Pull two roles out after the fact and place them elsewhere, and the boundaries of the surrounding categories shift with them, leaving a classification nobody can explain the origin of.

The workable order is the other way around. First the roles you have established yourself are set aside. They stay exactly as agreed and take no further part in the calculation. Only then does the model group the roles that remain.

The difference sounds like a detail, and it decides whether you can still retell the classification a year later.

Deviating is allowed, but it has to be retellable

Departing from the calculated outcome is not a weakness. There are good reasons for it, and most of them are things a model cannot know:

• The job description is outdated and no longer covers what the person actually does.
• Two roles differ on paper but rotate across the same tasks in practice.
• There is a historical agreement with the works council you do not want to walk away from unilaterally.
• The difference the model measures is real, but too small to justify a separate category.

What counts in all four cases is that the reason was written down at the moment the decision was made, not reconstructed afterwards. A classification whose order can no longer be traced stands weaker before a works council, a regulator or a court than one with a dated trail underneath it.

So we record every manual classification with the reason the employer gave and the date it happened, in a log nothing disappears from. A correction is a new line, because that is what a log is. And the moment the classification was established is recorded separately, so it stays visible what changed after that.

Whether that happened after the pay data arrived is the first thing a works council will ask. A classification that demonstrably came about before the figures were on the table is a different conversation from one that was adjusted afterwards.

What an instrument cannot do here

To be straight about the limit: we do not test a manual classification on its merits. Merge two roles that are substantively far apart and no alarm goes off. All that is recorded is that you decided it that way, and why.

That is a deliberate choice rather than a shortcoming. An instrument that overrules your agreement makes the agreement worthless. But it does mean the responsibility for a deviation sits with the employer and the works council, not with the software. Lean on it too often and you no longer have a method, you have a list of preferences.

A useful test: can you explain to someone who was not in the room why these two roles were classified together, without referring to what the system said? If you cannot, it is probably not an agreement but a habit.

How to explain this to your works council

The sentence that works worst in a works council meeting is "this is what the system said". It invites exactly the discussion you do not want, one about the system instead of about the work.

What does work is walking through the classification category by category with the reasoning attached: on which of the four statutory factors these roles resemble each other, where they differ, and why that difference was or was not large enough for a category of its own. The four factors and how to apply them are set out in we are covered by a collective agreement, does that settle pay transparency.

For the roles you classified differently together, that conversation is the shortest, because there the reasoning is the decision itself. That is not a weak spot in your file. It is the one part your works council is party to.

What you can do now

1. Agree who establishes the classification before you start weighing. It is a decision by the employer and the works council, not the output of a model.
2. Apply manual classifications before the calculation rather than after, so the rest of the classification does not shift around them.
3. Write down the reason for every deviation at the time, with a date.
4. Record when the classification was established, and keep track of what changed after that.
5. Walk the categories through with your works council on the reasoning, not on the outcome.

The bill is before the Dutch House of Representatives and has not been adopted; the intended entry into force is 1 January 2027.

Aycabtu provides the weighing, the categories and the documentation underneath. You establish the classification. This article is information with source references, not legal advice.