The Feedback Lie cover

The Feedback Lie

Why the most ritualised act of modern leadership quietly makes people worse, and what we are really protecting when we refuse to look at the evidence

Somewhere this week, in a glass-walled room booked for forty-five minutes, a manager is going to lean forward slightly, soften their voice, and say the words that every human being on earth has learned to dread: I want to give you some feedback. And the person sitting opposite is going to do something remarkable and entirely involuntary. Their heart rate will climb. Their cortisol will rise. Their attention will leave the work under discussion and travel inward, toward the far more urgent question of whether they are about to be told that they are not who they thought they were. Whatever the manager says next will land in a nervous system that has already stopped listening and started defending. And afterwards, in the debrief, HR will record that the conversation happened, tick the box, and count it as development.

We have built the entire architecture of modern people management on top of a practice whose evidence base, when you actually read it rather than quote it, is one of the most quietly embarrassing findings in organisational psychology. Not because feedback never works. It sometimes works beautifully. But because the average conceals a distribution that nobody in the industry wants to talk about, and because the conditions under which it fails are, with almost comic precision, the exact conditions we have spent thirty years designing into our performance cycles.

The number nobody quotes

In 1996, Avraham Kluger and Angelo DeNisi published in Psychological Bulletin what remains the most comprehensive meta-analysis of feedback interventions ever assembled: more than six hundred effect sizes drawn from decades of controlled studies. The headline was respectable. The average effect of feedback on performance was positive, around d = 0.41, which in the soft sciences is the kind of number that gets you a consulting practice.

Then there is the other number. In more than a third of those studies, feedback did not merely fail to improve performance. It reduced it. People who received feedback performed worse than people who received nothing at all.

Sit with that for a moment, because the reflex is to soften it. This is not “feedback delivered badly.” This is the intervention itself, studied under laboratory conditions, producing damage better than a third of the time. If a pharmaceutical company brought us a compound with that profile, we would not put it in the drinking water. We would print a warning on the box, restrict its indications, and insist on knowing exactly which patients it harms before we prescribed it to anyone. Instead we made it mandatory, scheduled it annually, trained managers to deliver more of it, bought software to increase its frequency, and called the resulting increase in volume a culture of continuous feedback.

The mechanism Kluger and DeNisi proposed is the part that should keep leaders awake. Feedback does not act on performance directly. It acts on attention. When the information lands close to the task, attention stays on the task, and behaviour improves. When the information lands on the self, attention moves to the self, and the self has more pressing business than learning. It has to defend its standing, negotiate its story, and work out what this means for its future. And if you audit the language of your own last performance cycle honestly, you will notice that almost none of it was about the task. You need more executive presence. You are not strategic enough. You come across as defensive. These are not statements about work. They are statements about a person’s worth, delivered by someone who controls their salary.

We engineered the perfect threat and then scheduled it

In 2004, Sally Dickerson and Margaret Kemeny published a synthesis of two hundred and eight laboratory studies of acute stress in the same journal, asking a simple question: which stressors actually move human physiology the most? The answer was unambiguous. The largest cortisol responses, and critically the slowest recoveries, came from tasks that combined two specific ingredients: uncontrollability and social-evaluative threat. Situations you cannot influence, judged by someone whose opinion of you matters.

Now describe the annual performance review without using its name. A person is assessed against criteria they did not set, by someone whose judgment they cannot overrule, in a conversation whose outcome determines their compensation, their status, and their future, and which has already been decided in a calibration meeting they were not invited to attend.

We did not accidentally create a stressful conversation. We reverse-engineered the single most reliable cortisol trigger in experimental psychology, put it in every manager’s calendar, and then built a training industry around the observation that employees “struggle to receive feedback well.”

The rating is a portrait of the rater

If the physiology were the only problem, we could fix it with better delivery, which is exactly what the coaching industry has been selling for two decades. The harder problem is that the content is largely noise.

In 2000, Scullen, Mount and Goff published in the Journal of Applied Psychology an analysis of the latent structure of performance ratings, and found that idiosyncratic rater effects accounted for 62% and 53% of the variance across their two samples, while the actual performance of the person being rated accounted for only 21% and 25%. The largest single thing measured by a performance rating is not the performance of the person being rated. It is the person holding the pen. Their standards, their memory, their mood, their relationship with the ratee, their private theory of what good looks like.

Which means that when you hand someone their rating, you are handing them a portrait of their manager and asking them to accept it as a portrait of themselves. And we do this with a straight face, in a process we describe as objective, supported by a nine-box grid whose main empirical achievement is a beautifully rendered map of our raters.

What people actually do with your feedback

Here is where it stops being a measurement problem and becomes a cultural one.

A team of researchers at Harvard Business School, led by Paul Green, followed roughly three hundred employees of a manufacturing company across four years of peer review data, looking at what happened to people’s working relationships after they received feedback that fell below their own view of themselves. The pattern was consistent and, once you see it, difficult to unsee. People who received disconfirming feedback from colleagues they were not obliged to work with were around 37% more likely to drop those relationships altogether. Where the critic could not be dropped, because they sat in the same unit, people compensated by reaching outward, building new and more distant ties with colleagues more likely to affirm the story they preferred about themselves. In a controlled follow-up, roughly 30% of participants abandoned their partner after disconfirming feedback, against fewer than 9% after confirming feedback.

And their performance the following year was worse.

Read that as an organisational designer rather than as a psychologist and the implication is devastating. Every honest review you deliver quietly rearranges your informal org chart in favour of flattery. Over a few cycles, the people who tell the truth get structurally isolated and the people who affirm get pulled closer, not through malice or politics but through the ordinary human machinery of self-protection. You did not build a feedback culture. You built a slow, self-reinforcing sorting mechanism that selects for agreement, and you paid a consultancy to install it.

Meanwhile the other side of the conversation is failing too. Since Rosen and Tesser named it in 1970, we have known about the MUM effect, the robust human reluctance to transmit unwelcome news. Managers do not soften negative feedback primarily to protect the receiver. They soften it to protect themselves from the discomfort of being its source. So the system in its entirety looks like this: raters whose ratings mostly describe themselves, delivering diluted information under conditions engineered to maximise threat, to people who will subsequently avoid them, with a measurable decline in performance to follow. And the metric the organisation reports to its board is the completion rate.

The lie is not that feedback fails

Around 2015, a wave of large companies announced they were abolishing performance ratings. Adobe, Deloitte, General Electric, and a long queue of imitators. It was covered as a revolution in humane management. A decade on, a 2025 report from Cornell’s Center for Advanced Human Resource Studies records the quiet reversal: among the companies that removed ratings, some found they could no longer connect performance assessment to compensation decisions and talent designation, and are now bringing ratings back in some form or considering it.

That reversal is the most honest document in this entire field, because of what it reveals about the original purpose. When the development story was removed, the thing that broke was not development. It was pay administration and legal defensibility. Nobody reported that learning collapsed. They reported that they could not allocate the bonus pool.

So here is the lie, and it is not the one you expected. The lie is not that feedback does not work. The lie is that we have spent thirty years describing a control system as a development system, and then blaming human beings for responding to it exactly as the control system it is. People are not bad at receiving feedback. People are excellent at detecting what a ritual is actually for. The defensiveness you find so frustrating in your direct reports is not a character flaw. It is accurate threat perception.

What survives the evidence

I am not going to oversell the alternatives, because the honest position is that the evidence against the status quo is far stronger than the evidence for anything we might replace it with. But three things survive contact with the research, and they are not difficult.

The first is separation. If the conversation determines money, it cannot also produce learning, because the physiology will not allow both to occupy the same room. Split them, by weeks if you can and by person if you dare, and stop pretending that the combined meeting is a kindness.

The second is direction. Kluger, the same researcher who exposed the problem, spent the second half of his career on what he called the feedforward interview, built on a deceptively simple sequence: ask someone to describe a specific moment when they were genuinely at their best, work backwards to decode the conditions that made it possible, and then ask whether their current plans reproduce any of those conditions. The evidence base here is thinner than I would like, and I will say so plainly rather than sell it to you. But it is directionally correct for the reason the meta-analysis already told us: it keeps attention on the work and away from the self.

The third is proximity. The only feedback with a consistently good track record is specific, close to the task, close in time to the event, and about what happened rather than about who someone is. The difference between “the second section of that proposal buried the commercial argument on page nine” and “you need to be more commercially minded” is not a difference in tone. It is the difference between an intervention that improves performance and one that has a better than one in three chance of degrading it.

Why this belongs to the experience conversation

The formula I keep returning to is HX = CX + EX. Human experience is not a department. It is the sum of what customers experience and what employees experience, and the two are not separable, because they are produced by the same people in the same nervous systems on the same day.

You cannot run an employee experience built on annual social-evaluative threat and then ask those same people to produce a customer experience built on generosity, presence, and the willingness to hear something uncomfortable without flinching. Whatever you do to people internally, they will do to customers externally, because that is the only model of being treated that your organisation has taught them. The organisation that punishes its people for defensiveness while institutionalising the conditions that produce it is not confused. It is simply not looking.

So before your next cycle opens, do this. Take the last ten pieces of feedback you personally delivered and sort them into two piles: the ones that described the work, and the ones that described the person. Then ask yourself, honestly, which pile you would have wanted to receive from someone who controls your salary.

Most leaders find the answer uncomfortable. That discomfort is the most useful piece of feedback you will get all year, and notice that nobody had to give it to you.

#TheH2HExperiment , #PerformanceManagement , #Leadership , #EmployeeExperience , #OrganizationalPsychology , #FutureOfWork

References

CEB. (2016, November 7). Performance reviews: Don’t remove the ratings [Press release]. Survey of nearly 10,000 employees across 18 countries.

Dickerson, S. S., & Kemeny, M. E. (2004). Acute stressors and cortisol responses: A theoretical integration and synthesis of laboratory research. Psychological Bulletin, 130(3), 355–391.

Green, P., Gino, F., & Staats, B. R. (2017). Shopping for confirmation: How disconfirming feedback shapes social networks. Harvard Business School Working Paper No. 18-028.

Heine, E. C. E., Stouten, J., & Liden, R. C. (2026). Performance feedback: A critical systematic review. Journal of Organizational Behavior, 47(2), 312–361.

Kluger, A. N., & DeNisi, A. (1996). The effects of feedback interventions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin, 119(2), 254–284.

Kluger, A. N., & Nir, D. (2010). The feedforward interview. Human Resource Management Review, 20(3), 235–246.

Rosen, S., & Tesser, A. (1970). On reluctance to communicate undesirable information: The MUM effect. Sociometry, 33(3), 253–263.

Rosen, S., & Tesser, A. (1972). Fear of negative evaluation and the reluctance to transmit bad news. Journal of Communication, 22(2), 124–141.

Scullen, S. E., Mount, M. K., & Goff, M. (2000). Understanding the latent structure of job performance ratings. Journal of Applied Psychology, 85(6), 956–970.

Link to original post: The Feedback Lie