Works Councils and Employee Research: How to Run Employee Studies in Germany and the EU

Answer first: in Germany and much of continental Europe, launching an employee survey without involving the works council can make the measure legally void, and the consent form you would use for customer research is the wrong instrument entirely — because employee consent is generally not considered freely given. Those are the two facts that surprise most product and people teams, usually a week before launch. The good news is that both problems have the same solution: design the study so it genuinely cannot identify individuals, then get that design blessed in a works agreement before you send a single invitation.

This guide covers the co-determination trigger, why consent fails, what a works agreement should contain, the technical design that satisfies both, and the sequence that turns a six-week negotiation into a two-week one.

Why an employee survey is a co-determination matter

Under the German Works Constitution Act (Betriebsverfassungsgesetz, BetrVG), a works council holds three tiers of rights: information and hearing rights on personnel actions, consultation on economic changes, and full co-determination on the social matters listed in §87. Employee research usually lands squarely in that third tier.

The critical provision is §87(1) No. 6, which gives the works council full co-determination over the introduction and use of technical devices designed to monitor the behaviour or performance of employees. German labour courts read this expansively: it is enough that the system is objectively suitable for monitoring, whether or not you intend to use it that way. A platform that records who responded, when, from where, and stores free-text answers is objectively suitable. §87(1) No. 1, on matters of order and conduct in the establishment, is frequently engaged too, and the 2021 Works Council Modernisation Act explicitly extended council involvement into artificial intelligence and mobile working.

Full co-determination means what it says. The employer cannot act unilaterally: the council must actively agree. If agreement cannot be reached, the matter goes to a conciliation board (Einigungsstelle), whose ruling replaces agreement. If you bypass the right, the measure can be legally void, and the council can seek an injunction — which in practice means the study stops, sometimes after you have already collected data you now cannot use.

Comparable structures exist elsewhere. The Netherlands gives works councils consent rights over personnel-data systems, France requires consultation of the CSE on measures affecting working conditions and monitoring, and multinational rollouts can additionally engage a European Works Council under the EWC Directive. The German analysis is the strictest and the most useful to design against.

Why employee consent is the wrong legal basis

Separately from labour law, the GDPR governs the data. The instinct is to collect consent. That instinct is wrong.

The European Data Protection Board has been consistent that, because of the inherent imbalance of power in an employment relationship, employee consent is unlikely to be freely given. Consent is only valid where a person can refuse or withdraw without detriment, and an employee asked by their employer to participate in a study is rarely in that position. Building your compliance case on consent means building it on a basis a regulator is predisposed to reject.

Article 88 GDPR allows member states to make more specific rules for employment-context processing, and several — Germany, France, the Netherlands — have done so. That is why employee research in Europe is a national-law question layered on top of an EU-law question, and why a single pan-European template rarely survives contact with local counsel.

The practical consequence: rely on a basis that fits the situation — typically legitimate interests for genuinely voluntary, aggregate-only research, or a legal obligation where a survey is required by other law — and put the safeguards in the works agreement instead of in a consent checkbox. Where consent does appear, it should be for genuinely optional extras (agreeing to a follow-up conversation, agreeing to be quoted), never for participation itself.

Anonymity is a technical requirement, not a promise

Here is the leverage point. Almost every objection a works council raises — monitoring, performance inference, retaliation risk, manager-level scrutiny — dissolves if the study is architecturally incapable of identifying individuals. Councils are not opposed to hearing from employees. They are opposed to a system that could be turned against them.

A design that survives scrutiny does all of the following:

The threshold rule is the one people underestimate. In a 900-person company, "Engineering, Munich, women, senior" is a group of four, and everyone in the room can name them. Suppression thresholds are what make an anonymity promise structurally true rather than merely sincere.

What belongs in the works agreement

A Betriebsvereinbarung covering employee research should be specific enough that the council does not have to trust you and general enough that you do not renegotiate every quarter. Cover:

1. Purpose — what you are trying to learn, and the explicit exclusion of performance assessment.
2. Scope and cadence — which populations, how often, and a cap on frequency so employees are not surveyed weekly.
3. Voluntariness — participation is voluntary and non-participation carries no consequence, stated in the invitation.
4. The tool — named platform, where data is hosted, the processor agreement, and sub-processors.
5. Data categories — exactly which fields are collected, including metadata such as timestamps and device information.
6. Anonymity architecture — the suppression threshold, the cross-tab restrictions, and the ban on individual-level export.
7. Access — who sees raw data, who sees aggregates, and how that is enforced and logged.
8. Verbatims — the de-identification process, and who performs it.
9. AI processing — if an AI conducts or analyses the interviews, say so plainly: what the model does, what it is not permitted to do, whether outputs feed any decision about an individual (they should not), and that a human reviews findings.
10. Retention and deletion — dates, not intentions.
11. Council access to results — councils routinely, and reasonably, want to see the same aggregate findings management sees.
12. Review and termination — how the agreement is revisited.

Item 9 is now the one that stalls negotiations. Councils have become alert to AI in the workplace, and the fastest way through is total specificity about what the AI does and a hard, written commitment that no output is ever attributed to or used against an individual.

Where Koji fits

Koji's design happens to line up well with what a works council needs, because the platform was built for research rather than for management reporting.

The AI interviewer conducts a genuine conversation and probes follow-up questions, which is what makes an aggregate-only study worth running at all — the depth that normally requires a named facilitator arrives without one. Nobody in HR sits in the room, and nobody needs to listen back to identify a voice, because the analysis is automated. In practice this is easier to defend than a traditional focus group or a manager-led round of one-to-ones, where the person hearing the answer is the person writing the review.

The six structured question types — open_ended, scale, single_choice, multiple_choice, ranking and yes_no — matter here for a specific reason: the five non-open types produce aggregate values by construction. A study whose backbone is scales, choices and rankings, with open-ended probing on the questions that genuinely need reasoning, gives you a report that is largely composed of distributions rather than quotable text. That is a materially easier artefact to agree an anonymity architecture around than a pile of free-text responses.

Team access uses owner, admin and member roles with workspace scoping, so "only these three named people can reach the study" is an enforceable statement rather than a policy aspiration. Combine that with an agreed suppression threshold applied at reporting time and you have most of the technical safeguards the agreement will ask for.

Be straightforward about the limits, too. Any system that collects responses holds metadata; the honest position with a works council is not "nothing is recorded," it is "here is exactly what is recorded, here is who can see it, here is when it is deleted, and here is why none of it can be resolved to a person in any output you or we receive."

The sequence that saves six weeks

Teams lose time by building the study first and consulting the council last. Reverse it.

The pilot in week 3–4 is the highest-leverage step in the list. Showing a council the real report — visibly free of anything that could identify anyone — resolves more objections than any amount of written assurance.

Two failure modes to avoid

Running a "quick pulse" outside the agreement. Small, informal surveys are exactly where co-determination gets bypassed, and they set a precedent that poisons the negotiation for the programme you actually care about. Bring the pulse inside the agreement's cadence clause instead.

Promising anonymity you cannot deliver. If the survey has 30 respondents and you report by team, you have not run an anonymous study, whatever the invitation said. The credibility cost of one identifiable finding is years long — every future study gets lower participation and more guarded answers. See anonymous employee research with AI interviews for the design detail.

Frequently asked questions

Do we need works council approval for every employee survey? In Germany, assume yes for anything using a technical system to collect employee responses, because §87(1) No. 6 BetrVG is triggered by a system's objective suitability for monitoring rather than by your intention. The efficient answer is one framework works agreement covering employee research generally, with a light notification step per study, rather than a fresh negotiation each time.

Can we rely on employee consent instead of a works agreement? No, for two separate reasons. Consent is a data protection concept and does not displace a labour-law co-determination right at all. And within data protection, the EDPB's position is that employee consent is unlikely to be freely given because of the power imbalance, so it is a weak basis even on its own terms. Use a works agreement plus an appropriate legal basis, and reserve consent for genuinely optional extras.

What happens if we launch without involving the works council? The measure can be treated as legally void, the council can seek an injunction to stop it, and you may be unable to use data already collected. Beyond the legal exposure, it is a relationship cost that makes every subsequent programme harder.

Does this apply outside Germany? The specific §87 mechanism is German, but the pattern is not. The Netherlands gives works councils consent rights over personnel-data systems, France requires CSE consultation on monitoring and working conditions, and multinational rollouts can engage a European Works Council. Design to the German standard and you will usually clear the others.

How small can a reporting group be before anonymity breaks? There is no single legal number, and the honest answer depends on how well colleagues know each other. A threshold of 5 is a common floor and 10 is safer for sensitive topics. What matters more than the number is preventing filter stacking, since combining three innocuous filters is how a group of 200 becomes a group of 3.

Can we use AI to conduct employee interviews in a co-determined workplace? Yes, and it is often easier to agree than a human-moderated alternative, because no colleague hears the answer. The conditions are specificity and restraint: describe exactly what the AI does, commit in writing that no output is attributed to or used against an individual, keep a human in the loop on findings, and disclose the AI's role to employees in the invitation.

Related Resources

• Structured Questions Guide — the six question types that make aggregate-only reporting practical
• Anonymous Employee Research with AI Interviews — the design that keeps an anonymity promise
• Koji for HR and People Teams — running employee research at scale
• Voice of the Employee — building a listening programme that drives change
• Employee AI Adoption Research — the study most likely to need a works agreement in 2026
• GDPR-Compliant AI User Research — the underlying data protection baseline
• Research Data Access Controls and Audit Trails — enforcing the access limits your agreement promises