Continuous feedback vs annual reviews: why rhythm beats format
Read more Feedback Culture
Every 360 review is biased, because every rater is. That is not a flaw you can engineer out with a better questionnaire or a smarter algorithm. Bias is baked into the moment a human forms an opinion about another human. So the honest question is not how to remove bias from a 360, which is impossible, but what you do with it: make it visible, or hide it. Most tools hide it, by handing you a single averaged score that looks objective and quietly launders six people’s biases into one clean-looking number.
I sell a 360 tool, so here’s the thing my industry doesn’t put on the landing page: none of us removes bias. What good process and honest tooling can do is dilute it and stop concealing it. That’s a smaller promise than “objective feedback,” and it’s the only honest one. This piece is the version I wish more vendors would write.
Key Takeaways
- Bias in 360 reviews is inherent, not a defect. Every rater is biased, so every review is. You cannot remove it, only make it visible or hide it.
- The single averaged score is the main way tools conceal bias. It takes six differently-slanted opinions and reports one number that looks objective and isn’t.
- The common biases are predictable: leniency and severity, halo, recency, similar-to-me, and reciprocity or collusion. A named list is what lets you read a report critically.
- You dilute bias with process (enough raters, real recent exposure, behavioural questions, good timing) and you expose the rest by reading the distribution and the written evidence, not the average.
- No tool removes bias. The honest ones refuse to hide it. That is the whole difference worth paying for.
Start with the uncomfortable premise, because everything else depends on it. A 360 gathers opinions from people, and people are biased. Not maliciously, usually. A peer likes you, or once clashed with you on a project, or remembers last week more vividly than last quarter, or unconsciously rates people who work the way they do a little higher. None of that is fixable. It is what having a perspective means.
So when a product implies its scores are objective, it’s making a claim it can’t support. The number is not objective. It’s an average of subjective inputs, which is a different thing wearing a lab coat. The moment you accept that bias is permanent, you stop chasing the impossible goal (an unbiased review) and start chasing the achievable one: a review that is honest about the bias it contains, dilutes what it can, and shows you the rest so you can judge for yourself.
That reframe is the entire piece. Hold onto it.
You don’t need the full academic taxonomy, but you do need names, because a bias you can name is one you can spot in a report. These are the ones that turn up again and again.
Leniency and severity. Some raters mark everyone high; some mark everyone low. Neither is telling you about the person being reviewed. They are telling you about the rater’s default setting.
The halo effect. One vivid strength, or one memorable weakness, bleeds into every other answer. A brilliant communicator gets rated high on judgment, planning, and reliability too, whether or not the evidence is there.
Recency bias. The last month drowns out the ten before it. A strong recent project, or a recent stumble, gets weighted as if it were the whole period.
Similar-to-me bias. People rate colleagues who think, work, or communicate like they do more generously. It is quiet and almost entirely unconscious, and it is one reason feedback can quietly penalise people who work differently from the majority.
Reciprocity and collusion. In peer feedback especially, “I’ll rate you well if you rate me well” is a real dynamic, sometimes explicit, usually not. It inflates everyone and flattens the signal.
Underneath all of these sits the finding that unsettles people most: a large share of any single rating reflects the rater, not the person being rated. That is not an argument against 360s. It is the argument for them, because the fix for one biased perspective is more perspectives, which is the whole idea. It is also the argument against ever trusting a lone number. (Worth disambiguating one term: this is bias in workplace peer feedback, not the “peer review bias” of academic journals, which is a different conversation entirely.)
Here is where the tooling turns from neutral to actively unhelpful. Most 360 products take those six slanted opinions and do the one thing guaranteed to hide the bias: they average them into a single score per competency, and they put that number front and centre.
An average is a bias-laundering machine. It takes a leniency-inflated 5, a severity-deflated 2, and four honest 4s, and reports a calm, official-looking 3.8. The 3.8 tells you nothing true. It erases the fact that one rater is scoring on a different planet, that the spread is enormous, that there might be two camps who see this person completely differently. A number that clean, built from inputs that messy, is not a summary. It is a cover-up.
Imagine a manager opening a report that says “Collaboration: 3.6 out of 5.” Reassuringly mid. What the average hides is that three reviewers rated it a 5 and two rated it a 2, because the person collaborates beautifully with the design team and barely talks to engineering. The truth was a split, and a specific, actionable one. The average turned it into a shrug. This is the same reason a single averaged number on its own is not feedback: I have made the full case against standalone rating scales separately, and it applies with double force once you add bias to the picture.
To be clear, this is a category-level habit, not a knock on any one product. Plenty of tools default to the average because it looks tidy and demos well. The point is that tidy is exactly the problem when the underlying reality is a spread of biased opinions.
If you can’t remove bias, the next best thing is to refuse to hide it. That means showing three things the average destroys.
The distribution, so a split reads as a split. Five people rating someone 3.6 on average is a very different situation depending on whether everyone said 3.6 or half said 5 and half said 2, and you should be able to see which at a glance.
The written evidence behind each rating, kept attached to it, so you can tell a leniency-inflated 5 (“great, love working with them”) from an earned one (“caught the pricing error before launch, twice”). Numbers hide bias; the sentences next to them expose it.
The self-versus-others gap, because the distance between how someone rates themselves and how everyone else does is often the most honest signal in the whole report.
This is the one place a tool genuinely helps, and it is worth being precise about how, because the honest framing matters. It does not remove bias. What it can do is show the distribution and keep every rater’s words attached to their ratings, so you’re reading the spread and the evidence rather than a laundered average. That is how we built the report in Lynxify.me: the number never travels alone. But you can enforce the same discipline in a spreadsheet. The principle is what matters: never let an average be the last word.
Tooling handles visibility. Diluting the bias in the first place is a process job, and most of it costs nothing.
Include enough raters. The single most effective bias control is numbers. One biased voice in a pool of three is a third of the signal; the same voice in a pool of six is diluted. This is who you pick as raters, and it matters more than any feature: choose people with real, recent, direct exposure to the person, and enough of them that no single perspective dominates.
Ask for behaviour, not traits. “Rate their leadership, 1 to 5” invites every bias at once, because it is asking for a gut impression. “Describe a recent moment when their decision changed the team’s direction” asks for an observed event, which is much harder to halo or fabricate. Shifting from trait ratings to observable behaviour and specific examples is the question-design move that starves halo and leniency bias of their oxygen.
Run it close to the events. Recency bias thrives on long gaps, because when people cannot remember the details they fall back on a general impression, and general impressions are where bias lives. Feedback gathered while the examples are fresh has less room to drift.
Keep it developmental, and use anonymity deliberately. When a 360 is tied to pay or promotion, reciprocity and self-protection spike, because now there is something to game. Developmental feedback has less incentive to inflate. And anonymity, used where it genuinely lifts candour, can reduce the collusion dynamic, though it is not automatic and carries its own tradeoffs at small scale.
Picture a People Ops lead at a 50-person company who does all of this: six raters instead of three, example-based questions instead of trait scores, run a couple of times a year rather than once, kept away from comp. The bias does not vanish. But no single skewed voice can hijack the result, the halo has less to grab, and what bias remains is sitting in plain sight in the written comments where a careful reader can weigh it. That is a realistic best case, and it is a very good one.
Stop trying to buy your way to objectivity. It is not for sale, and the products implying otherwise are selling a tidy number that hides exactly the thing you most need to see. Every 360 is biased. The only choice you actually have is whether that bias is visible and diluted, or invisible and concentrated behind a decimal point.
So build for honesty, not objectivity. Enough raters to dilute any one of them, questions that ask for evidence, timing that respects memory, and a report that shows you the spread and the sentences instead of an average that pretends the spread away. A messy distribution you can actually read beats a clean number you should not trust, every time. If you want to run one that keeps the evidence attached to the ratings by default, you can start a 360 in Lynxify.me. But the discipline matters more than the tool: never let the average have the last word.
Dmytro Shtapauk
Practical tips and experience from top-notch experts
