A selection procedure ends on the day the decision is made. For the documentation, something else begins on that day: a deadline most organisations do not have in their calendar, because it only becomes visible once it has passed.
The window
The German General Equal Treatment Act (AGG) gives rejected candidates two routes, and each has its own clock. Under § 15 (4) AGG a claim has to be asserted in writing within two months, counted from receipt of the rejection. Once that has happened, § 61b (1) of the Labour Court Act allows a further three months for the action itself.
In practice that means up to five months can pass between the rejection and the moment an organisation first has to explain in court how the decision came about. The rejection went out in March, the claim is served in July. Anyone who can still produce the observer sheets, the requirement analysis and the course of the procedure at that point is in a different position from someone relying on memory.
This is not legal advice, and the individual case belongs in the hands of lawyers. For organising a selection procedure, though, the basic arithmetic is enough: the documentation has to outlast the procedure.
Why the burden of proof shapes the documentation
The point at which documentation turns from an administrative act into protection is § 22 AGG. Anyone who presents indications suggesting discrimination shifts the burden of proof. The employer then has to show that the prohibitions on discrimination were not breached.
That construction is the real reason structured procedures stand up better in a defence than unstructured ones. Not because they produce better decisions, although as a rule they do. But because they make a decision explainable: there was a requirement analysis before the procedure, every candidate worked on the same tasks in the same order, several observers assessed independently of one another, the assessment followed defined criteria, and the decision can be traced back to those criteria.
A procedure that runs slightly differently at every site can produce the same outcome. It is simply harder to evidence.
What that means in operation
The typical case from practice looks like this. An organisation runs assessments at four sites. The procedures are professionally set up, the observers trained, the assessment sheets uniform. What is missing is the common place.
The sheets sit as scans in a folder on the site’s drive. The invitations went out from Outlook. Whoever stepped in as an observer that day because someone was ill appears in no system, only in the coordinator’s memory. The requirement analysis is a Word document from 2023 that has been revised twice since, and nobody can say which version applied in March.
When the claim then arrives, a search begins. In our experience from two decades of running these procedures, that search costs more time than the procedure itself, and it often ends with a gap remaining. Not because anyone was careless, but because nobody in March knew that somebody would ask in July.
The conflict that has to be named
At the same time the GDPR demands the opposite. Personal data may only be stored for as long as the purpose requires. Keeping candidate data indefinitely after a rejection is not an option.
Both requirements can be reconciled, but not by instinct. They need a decision: which documents are kept for how long, from what point are they anonymised, when are they deleted, and who may look into them in the meantime. The retention horizon usually follows the deadlines above plus a buffer. Access rights are independent of that: the fact that a file still exists does not mean everyone may see it.
Make that decision once and record it in the system, and it does not have to be made again for every procedure. Fail to make it, and it gets made anyway — uncontrolled, and in both wrong directions at once: some folders are cleared out after four weeks, others have been sitting on a drive for six years.
What flynne contributes
Our answer to this problem is unspectacular: the documentation comes into being while the procedure runs, not afterwards. Because every procedure grows out of the same template, it follows the same course at every site. Every candidate has a complete history with documents and notes attached to it. Every action is logged in an audit-proof way, with time, decision and the person who carried it out. Whoever stepped in as a replacement at short notice appears in it, because the rescheduling happened in the system and not over the phone.
The deadlines are stored rather than remembered. Six months after a procedure ends, candidate data is anonymised; two years later it is deleted in full. Visibility periods limit access before that, regardless of how long the data still exists.
None of this replaces a legal review or good diagnostics. It makes sure the question in July can be answered without anyone having had to think of it in March.
