How to roll out 360 feedback for the first time
Read more 360 Reviews
A rating scale on its own is not feedback. A 3.4 out of 5 on “communication” tells you where someone supposedly stands and nothing about what to do on Monday, and often it is not even measuring what it claims to.
I build 360 feedback software that uses rating scales, so this is not a purity argument to throw the numbers out. It is the opposite. The problem is not that 360 feedback uses rating scales; it is what happens when the number becomes the feedback, which is the default almost everywhere. Used that way, a rating scale corrupts a 360 in a few predictable ways, and none of them are fixed by adding more scale points or a nicer dashboard.
Here is the case against the standalone number, and the alternative I would actually run: demote the rating, anchor it to behavior, and let the words carry the weight.
Key Takeaways
- A rating scale on its own is a verdict without evidence. It says where someone stands, not what to change, so it produces no action.
- Averaging ratings across raters describes nobody. A person scored 5, 5, 5, 1, 1 averages to 3.4, which erases the actual story; the spread is the signal, not the mean.
- Rater bias (central tendency, leniency, the halo effect) compresses the scale into noise, and averaging bakes the distortion into the final score.
- Numbers invite ranking, calibration, and pay decisions, which turns a developmental 360 into an evaluation and kills honest answers.
- The fix is not to delete ratings but to demote them: anchor each to observable behavior, pair it with a written answer, and read the distribution instead of the average.
Start with what a rating actually is. When you ask someone to score a colleague 4 out of 5 on leadership, you are asking for a global judgment about a trait, delivered as a number. The number looks like a measurement. It is an impression wearing a lab coat.
The tell is that you cannot act on it. “You are a 3 on delegation” gives the person nowhere to go. A 3 relative to what, improved how, by doing what differently next week? The score marks a position on a line nobody drew and hands back no instructions. Compare that to a single written sentence: “you tend to keep the interesting parts of a project and hand off the rote parts, so your seniors feel stretched and your juniors feel like assistants.” One of those is feedback. The other is a grade.
Imagine a manager who opens her 360 report and reads that eight people scored her 4.1 on leadership. She has no idea what to do with that on Monday. Now imagine the same report told her that four people used the word “decisive” and three independently said she cuts people off in meetings. Same round, same raters, only one version changes her behavior. This is the aside in how to run your first 360 review that deserves its own argument: four open questions beat a forty-item rating scale, and this is why.
Here is where the standalone number stops being merely useless and starts being misleading. A 360 gathers several raters, so at some point the tool averages their scores and shows you one figure per criterion. That average is where the story goes to die.
Picture a person reviewed by five colleagues on collaboration, scored 5, 5, 5, 1, 1. The report shows 3.4. There is no one who thinks this person is a 3.4. Three people think they are excellent and two think they are a real problem, which is a fascinating and important thing to know, and the average erases it completely. The reviewee reads 3.4, shrugs, and files it under “fine, I guess.”
The spread is the signal. Disagreement between raters is the most interesting thing a 360 can surface, because it usually means the person behaves differently with different groups, or that one group sees something the others are missing. A number that averages the disagreement away throws out the exact insight you ran the review to find.
Even before you average, the raw scores are shakier than they look, because people are bad at turning judgments into numbers in consistent ways. This is not a knock on your team; it is well-worn ground in how ratings behave.
Some raters sit on the middle of the scale to avoid conflict, so everyone lands a 3 or a 4, a pattern known as central tendency. Others mark high across the board because they like the person or do not want to seem harsh, which is leniency. And the halo effect means someone who is genuinely strong in one visible area tends to get high marks on everything, including things the rater never actually observed.
Each of these quietly compresses a five-point scale into a two-point scale of “good” and “great,” and then averaging bakes the distortion in. The final number is a faithful measurement of your raters’ scoring habits and only loosely related to the person being reviewed. Written answers are not immune to bias either, but a sentence carries its own evidence: you can read “she rewrote my draft without telling me” and judge it for yourself. A 2.5 gives you nothing to check. This is also why question design matters more than scale design, and why 360 feedback questions that surface blind spots get people describing what they saw instead of rating who someone is.
There is a subtler corruption, and it is the one I would lose sleep over. Numbers invite comparison. The moment feedback comes back as scores, someone senior will want to line the scores up, rank people, feed them into calibration, and eventually attach them to pay. It is almost gravitational.
The trouble is that a 360 only works when people believe it is developmental. The instant raters suspect their scores decide someone’s raise or ranking, they stop being honest and start being strategic, and you are back to the dishonesty that makes the annual review useless. Written feedback resists this pull, a little, because prose is harder to average into a stack rank. A column of numbers is built for exactly that.
Picture a team that introduces 360s purely for development, with a clean 1-to-5 scale because it seemed tidy. Two rounds later, someone has exported the scores into a spreadsheet, sorted them, and used them to argue a promotion case. Nobody decided to turn the developmental 360 into a ranking engine. The numbers just made it the path of least resistance.
So here is what I would actually run, and it still uses ratings. It is a 360 where the rating is demoted from verdict to pointer.
Lead with the written question. Ask people to describe observable behavior in a specific situation rather than score a trait: “what is one thing this person should keep doing, with an example” beats “rate this person on execution.” The words are the feedback; that is where the signal and the actionability live.
Keep a rating if you want one, but anchor it and pair it. A number is genuinely good for one thing, spotting a pattern fast across a lot of responses, so attach the scale to a concrete, observable criterion and require a written “why” next to every score. That way the number is always a pointer to a sentence rather than a claim of its own. A 2 with an example is data. A 2 on its own is a shrug.
Read the distribution, not the average. Look at how the scores spread before you look at what they average to. Agreement tells you what is settled; disagreement tells you where to read the comments closely. That is the reverse of how most dashboards are built, which lead with the mean and bury the spread.
And keep it developmental on purpose. Do not rank on it, do not calibrate on it, do not wire it to compensation, for all the reasons above.
This is, not coincidentally, how we built written feedback and criteria ratings together in Lynxify.me: every rating sits next to the written answer that explains it, and the report shows the spread across raters rather than flattening it into one number. I mention it because it is the model I am arguing for, not because you need the product to run it. You can do all of this in a shared document with a single rule that no score travels without a sentence.
Almost every corrupted 360 I have seen got that backwards. It treated the score as the answer and the comments as optional colour, and then wondered why the whole exercise felt hollow and nobody changed anything afterward.
Flip it. Build your 360 so the writing carries the weight and the rating just tells you where to look, read the spread instead of the mean, and keep the entire thing pointed at development rather than ranking. Do that and the number stops corrupting the feedback and starts doing the one modest job it is actually good at.
If you want to run it that way without a spreadsheet that no score can escape, you can start your first 360 in Lynxify.me. But the rule matters more than the tool: no number travels without a sentence.
Should 360 feedback use rating scales?
Yes, but not on their own. A rating scale is useful for spotting a pattern quickly across many responses, and useless as the actual feedback, because a score tells someone where they stand and not what to change. Use ratings only alongside written answers, anchor each score to a specific observable behavior, and never let a number stand as feedback by itself.
What should I use instead of a rating scale in a 360?
Not instead, alongside. Lead with open written questions about observable behavior in specific situations, and treat any rating as a pointer to those written answers rather than the answer itself. If you keep scores, require a written reason next to every one and read the spread across raters rather than the average, which is where the real signal lives.
How many points should a 360 rating scale have?
Far less than it seems. Teams agonise over 5-point versus 7-point scales, but the number of points barely matters next to two things that do: whether every score is paired with a written explanation, and whether you read the distribution instead of the average. Get those two right and a simple scale is fine; ignore them and no scale size will save you.
Dmytro Shtapauk
Practical tips and experience from top-notch experts
