Is 360° feedback anonymous? Anonymous in the answers, confidential in participation

Anonymous or confidential? What 360° feedback actually protects

The question “is 360° feedback anonymous?” has two answers, and a serious programme gives both. The individual answers are anonymous: nobody – not the leader, not HR – can see which rating a particular person gave. Participation is merely confidential: the leader usually knows who was invited, because in most programmes they nominated their raters themselves, and often sees who has completed the questionnaire. Anyone who promises raters “full anonymity” is promising something the process cannot deliver.

DefinitionAnonymity threshold

The minimum number of completed responses a rater group (supervisors, peers, direct reports) must reach before its average scores are shown separately in the report. Below it, the group’s answers are folded into a pooled “all others” view or not shown at all.

The table describes the target state. Whether your programme reaches it depends on three mechanisms: the anonymity threshold, the handling of supervisors, and the rules for written comments. Who gets to open the finished report is a separate question – we answer it in Can HR see individual 360° results?.

The anonymity threshold: why two is a floor, not a target

360° feedback earns its value from separating perspectives: Bracken, Rose and Church (2016) list formally segmented rater groups among the conditions for a process to count as 360° feedback at all, because a supervisor observing something different from the team is precisely the insight. That same separation creates the anonymity problem – the smaller the group, the closer its average sits to one person’s answer.

With a single response, the group average is that person’s answer – a threshold of one is no threshold. From two responses onward, the average stops belonging to any individual. That is why two is the floor a system must enforce as a minimum. It is not yet a good number:

• One voice in two is thin cover. If a leader has two direct reports and sees a low group average, they know it came from two people they know by name – anonymous on paper, not in the room.
• One drop-out collapses the group. With exactly two nominees, a single decline means the group can no longer be shown separately.
• Two answers are barely reliable. Research on the reliability of 360° ratings keeps reaching the same conclusion: more raters are needed than are typically invited (Greguras and Robie, 1998). Nowack and Mashihi (2012) note that rater groups with two or fewer responses may be inadequate for reliable measurement, and Hensel and colleagues (2010) found that two or three peer raters produce low reliability, with roughly six to ten raters needed, depending on the trait, to reach a reliability of 0.7.

The practical consequence: treat the threshold as a floor, invite noticeably more people than it requires, and choose the raters deliberately. We work out how many per group makes sense in How many raters does a 360° need?.

The subtraction problem: why visible groups can expose a hidden one

A per-group threshold is not enough if the report also shows an overall average across all raters. Once every group but one is shown separately, the missing group can be calculated back from the total – even though on its own it sits below the threshold.

Why supervisors are the exception – and who must know before they answer

Most leaders have exactly one supervisor. With a threshold of two, that supervisor’s feedback would never appear separately; it would always disappear into “all others”. Many programmes solve this with a deliberate exception: the supervisor’s feedback is shown as its own column, on the grounds that supervisors are not anonymous anyway, usually give their feedback openly, and the leader needs it for the development conversation.

Both designs are defensible. What is not defensible is deciding after the survey: whoever answers as a supervisor must know beforehand whether their scores will appear on their own. In LEADBeyond 360°, supervisors follow the same threshold as every other group by default; the consulting team can enable an explicitly labelled exception for an individual assessment that shows the supervisor column even with a single response. Even then, the subtraction rule still applies: if the rest of the rater mix is too thin, the column folds back into “all others”.

Written comments: the real weak point

Thresholds protect numbers, not language. A comment gives its author away through writing style, through the language it is written in, and above all through content: “in the steering meeting on Tuesday” or “when project X escalated” can only be written by someone who was there. A system can do three things – and should leave a fourth alone:

• Never attribute a comment to a person, and display it without a name.
• Show comments in random order, so that sequence reveals nothing about group or timing.
• Show a comment’s rater-group label only when that group meets the threshold – and withhold all comments if only one person responded in total.
• No rewriting. No system should smooth or summarise comments to disguise authorship – the content would be lost, and AI rephrasing opens a data-protection question of its own.

The rest is process. Raters should describe patterns rather than incidents (“cuts discussions short” rather than “interrupted Ms M. on Tuesday”), avoid dates, names and project names, and write what they would be willing to say to the leader in a team meeting. And the leader should not try to attribute comments at all – a facilitated debrief keeps attention on the pattern behind the voices, not on the voice. The full set of rules for comments within a programme’s confidentiality architecture is covered in Confidentiality and data protection.

What to promise raters – and what not

Ten questions to ask your vendor

1. What is the default anonymity threshold, and who can change it – HR, the consulting team, or nobody?
2. Where is the threshold applied: when results are calculated, or only when they are displayed? Is there an export, an interface or an admin view that shows raw data per person?
3. What happens to groups below the threshold – are they folded into “all others” or do they disappear?
4. Is an overall average withheld when a small group could be calculated back from it?
5. How is supervisor feedback handled, and are supervisors told before they answer?
6. Are comments shown without names, in random order and without a group label below the threshold? What happens if only one person responded?
7. Are comments summarised or rephrased by AI – and if so, by which provider, where, and on which legal basis?
8. What does HR see: progress and completion rates, or scores as well?
9. What does the leader see about who was invited and who has completed?
10. What happens when responses arrive after the report has been generated – is the report silently changed, or is a new version created?

These questions belong in every tender – alongside the other criteria in our 360° buyer checklist.

How LEADBeyond 360° enforces anonymity

As a concrete example of what the answers to those questions can look like, here is our own system – not as the benchmark, but as a reference point for your vendor conversations.

By default, a report is generated only once the self-assessment is complete, at least two rater groups meet the threshold, and either the deadline has passed or every invited rater has finished. Only the consulting team can deviate from that, and every deviation is logged.

Two things LEADBeyond 360° deliberately does not do. It does not hide the rater list: the leader sees who was nominated and who has completed – because they usually nominated their raters themselves, and a hidden list would only fake an anonymity that does not exist. And comments are not rewritten; the responsibility for describing patterns rather than incidents stays with the raters, who can also decline an assignment. Raters never see a report, and HR sees progress but no scores – the details are in Can HR see individual 360° results?.