360 Feedback

For a long time the evidence about a person came from one manager's field of view. What a manager sees is output and status reports. That covers a lot, and it leaves specific gaps. A manager who joins a meeting for its last ten minutes sees a decision, not how the room arrived at one. How someone handled another team's urgent request, whether they wait until a colleague finishes a sentence, is known only to the people who worked beside them.

The upward direction was missing too. With no channel for input about a manager, problems in that seat only surface in some other shape. Three people leave one team in a year and each exit interview names a different personal reason. That team lead's review document contains a goal attainment percentage and nothing else.

When five of six team members write down the same thing, that is not individual dissatisfaction; it's a problem with how the team is being run. And with nowhere for those observations to collect, the company learns it only as a turnover number.

360 feedback exists to fill both gaps by adding directions. Peers, direct reports, and sometimes another team that collaborated all contribute. Including a seat that evaluates the manager is what most distinguishes it from the other processes in this branch.

Decide what to ask, choose the respondents, collect, summarize and deliver, then convert it into work for the next quarter. Collection runs one to two weeks, and summarizing and delivering takes another.

1. Set the questions. Keep it to five to eight, and ask about behavior. Asking whether someone collaborates well returns that they do. Asking what happened when opinions split, or what they handled first when an urgent request arrived, returns scenes.
2. Choose respondents. One manager, three or four peers who worked with them, and for a management role two or three of their reports. A self-assessment on the same questions surfaces the gap in perception. The person proposes the list and the manager checks once that these are people they actually worked with.
3. Decide the anonymity boundary and say so in advance. Grouping peers and reports as anonymous while the manager's input is named is the common arrangement. Anonymity needs at least three people in a group; with two, the sentences alone reveal who wrote them. People want the boundary stated before they write rather than after, because how candid the answer is depends on it.
4. Collect inside a fixed window. Leaving it open beyond two weeks doesn't raise the response rate, it just moves everything to the deadline.
5. Summarize before delivering. Raw responses don't get handed over as-is; points several people raised get grouped, and anything one person raised alone is marked as such. Delivery happens in a session with the manager or the facilitator rather than alone.
6. Convert it into one or two pieces of work for the next quarter. Stop short of this and the whole process produces one document that gets read once. One item is enough, and it should be something the next cycle can actually verify.

Most companies keep the result out of rating and bonus calculations. Connect it and respondents drift toward writing well of each other, which leaves nothing worth reading.

There are five kinds of participant: the subject, their manager, peers who worked with them, the people who report to them, and a facilitator who owns the questions, the schedule, and the anonymity handling. The fourth exists only when the subject manages people, and that's usually where the value of the process comes from.

The timing attaches to the performance review cycle. It has to finish before review documents get written to be usable as input, so starting collection two weeks before self-assessments open is a common schedule. When results reach the calibration meeting, reviewers can argue from evidence that isn't just what one manager saw.

Frequency is once or twice a year. Run it quarterly and each person writes eight questions about five or more colleagues every time, and by the third cycle the answers get short. Annual is the default at most companies for exactly that reason.

The subject pool isn't everyone either. Ten people writing about each other produces ninety responses, which nobody finishes reading. Narrowing to management seats and people who work across several teams is the better call.

One condition applies. The process only produces value on teams where feedback already circulates. Say nothing all period and then raise a problem for the first time in this document, and it isn't a process, it's a notification.

"This cycle we're only running 360s for the team lead roles."

Narrowing the pool to hold the response burden down.

"Only two reports responded, so anonymity doesn't hold. I'm dropping that group."

Below the minimum, anonymity isn't real.

"Five of six wrote the same thing. That isn't one person's opinion."

Reading a repeated point as a signal rather than as a complaint.

Pagedeck ran it first for its four team leads. In Theo's results, four of five reports raised the same point: they learned the reasoning behind a decision only afterward. Theo's self-assessment listed decision speed as a strength, and the decisions were in fact fast. What nobody else had was how any of them were reached.

Mia set up a session to read the results together, organized by question. The work Theo took into the next quarter was one thing: leaving two lines of reasoning in the team channel whenever a decision gets made. Grace brought the results into the calibration meeting but kept them out of the rating calculation.

The most common reading is that anonymity produces candor. It's true that people write more when their name isn't attached. But anonymity is the minimum condition, not a sufficient one. What people actually read is what happened to the last thing they wrote. If the previous cycle's input changed nothing, this cycle comes back in safe sentences. Left uncorrected, the process runs every year and the results look the same every year.

The other reading is that collecting more makes it more accurate. The name, with its implication of more directions, invites that. But ten respondents times eight questions is eighty answers, and only some of them get read. A subject handed an unsorted pile of raw text remembers the single line that hurt most. Cutting to three or five respondents and five questions leaves more behind. Fewer questions read carefully change more behavior than many questions skimmed once. Left uncorrected, the summarizing work grows and no behavior changes.