Over a Third of Feedback Backfires. The Fix Is Aim, Not Volume.

Most leadership advice assumes that more feedback is better. The largest study ever conducted on the question found that over a third of feedback interventions made performance worse. Whether feedback helps or harms turns on a single design variable, and most organizations never look at it.

In our work with leadership teams, one request comes up more reliably than almost any other: “Help us build a stronger feedback culture.” The assumption underneath is that feedback is like a vitamin, where the only possible problem is deficiency. It is worth knowing that this assumption was tested, comprehensively, thirty years ago, and it failed.

Two researchers added up a century of evidence, and the average hid the real story

In 1996, Avraham Kluger and Angelo DeNisi published a monumental meta-analysis in Psychological Bulletin. They gathered nearly a century of research on feedback interventions: 131 usable studies, 607 effect sizes, more than 12,000 participants. On average, feedback improved performance moderately (an effect size of about 0.4). That average, however, concealed the finding the paper is remembered for: in over 38% of cases, feedback made performance worse. Not neutral. Worse.

A result like that cannot be explained by feedback quality alone, and the authors’ explanation became known as Feedback Intervention Theory. Its core claim is disarmingly simple: feedback works through attention, and attention is finite. Any piece of feedback directs that limited attention to one of three levels. It can point at the task and how to master it (what exactly went wrong in the analysis, what a correct version looks like). It can point at motivation around the task (progress against a goal, the gap between here and the standard). Or it can point at the self (what this result says about me, my talent, my standing). Their central conclusion, in their words: feedback intervention effectiveness “decreases as attention moves up the hierarchy closer to the self and away from the task.”

When feedback pulls attention to the self, cognitive resources leave the work and go into defending, comparing and ruminating. The task gets worse while the person processes the verdict on their identity.

What to learn: the aim of feedback matters more than the amount

Three implications hold up well after three decades.

First, feedback is an intervention with side effects, not a vitamin. Before increasing the dose, audit the aim. A leadership team that wants better performance from feedback should ask where its current feedback sends attention, because volume without aim simply produces more of whatever is already happening.

Second, the strongest deflectors toward the self are the moves that feel most generous. Praise, in the study, tended to attenuate feedback’s effect on performance: “you’re a natural at this” points attention squarely at identity, not at the work. So do rankings and normative comparisons, which convert a task question into a status question. The beloved compliment sandwich manages to deflect attention twice before the actual message arrives. None of this means recognition has no place; it means recognition and performance feedback are different instruments doing different jobs, and mixing them blunts both.

Third, feedback without a standard is drift. The interventions that reliably helped were those that kept attention on the task and its gap: information about the correct solution, cues about progress, feedback coupled to explicit goal setting. If there is no shared standard for what good looks like, feedback degenerates into opinion exchange, and opinion about work slides quickly into opinion about people.

How to learn it: audit where your feedback actually points

Here is where we take a position. You do not fix feedback with a training on feedback models. The material this discipline works on is your organization’s live, running feedback, so that is where it has to be learned, which is how we think most management capability is best built.

A practical loop for a leadership team:

1. Collect the evidence. Each leader gathers the last ten pieces of performance feedback they actually gave: review comments, 1:1 notes, project debriefs, the email written after the failed pitch. Real text, not remembered intentions.
2. Code the aim. For each piece, one question: where does this send the receiver’s attention, to the task and how to close the gap, to progress against a goal, or to the self? Most teams find the sobering pattern within twenty minutes.
3. Rewrite the deflectors. Take the self-pointing items and rewrite them so they name the work, the standard and the gap. “You need to be more strategic” becomes “the recommendation section argues from cost, but the board’s question was market position; next time, lead with that.”
4. Run one real cycle. Use the rewritten register in the next two weeks of actual reviews and 1:1s, with the standard made explicit before the feedback is given.
5. Reconvene and compare. What changed in the conversations, and in the work? Then repeat the loop once more.

When should you run this? If your organization has ritualized feedback (regular reviews, retros, surveys asking for “more feedback”) while performance stays flat, run the audit now; two cycles over about six weeks are usually enough to shift the register of a leadership team. If feedback in your organization is rare as well as badly aimed, start the loop anyway, but expect the first collection round to be thin, which is itself the finding.

For readers who use Management Kits, this is not a new philosophy but the design logic of the platform’s own feedback resources. The working topics under our Feedback & Reflection theme make the three commitments the meta-analysis rewards. A feedback worksheet structures each piece of feedback around context, observed behavior, impact and actionable next step, which keeps attention on the work. Feedback is tied to an explicit leadership learning agenda, so a standard sits behind every conversation. And feedback runs in loops of action, feedback, reflection and adjustment, given close to the action and led back to practice quickly. Structure, standard, fast path back to the task: thirty years on, the paper reads like the rationale for exactly these choices.

AI has made feedback abundant, which makes aim the whole game

The feedback bottleneck used to be scarcity: managers had limited time, limited observations, limited appetite for difficult conversations. That constraint is dissolving. AI tools now critique any draft in seconds, dashboards score calls and code and customer interactions continuously, and copilots volunteer commentary nobody asked for. Organizations are moving from a feedback desert to a feedback flood.

Kluger and DeNisi predict exactly what will separate the winners here, because the theory does not care whether the feedback comes from a person or a model. Machine feedback that points at the task and the gap, such as a concrete critique of an argument with a stronger version attached, sits at the productive bottom of the hierarchy and can accelerate learning dramatically. Machine feedback that points at the self, such as leaderboards, percentile ranks and automated performance scores delivered without a task path, industrializes precisely the kind of intervention the meta-analysis found most likely to backfire. The same applies to managers who let a model draft their appraisals: the fluent, praise-padded text these tools produce by default is aimed at the self almost by design. With AI already unsettling what expertise and seniority mean, the leadership discipline is no longer generating feedback; it is curating where the organization’s attention lands.

Where the theory ends

The limits deserve naming. Kluger and DeNisi themselves called their theory preliminary; much of the underlying evidence comes from measurable laboratory tasks, and thirty years of subsequent research have refined the picture, including on when praise helps and how feedback seekers differ from receivers. The framework is a lens, not a law. It will not tell you what the standard for good work should be; it only insists you have one, and that your feedback points at it.

Which brings us back to the feedback culture. The ambition is right; the noun is wrong. What the evidence supports is not a culture where feedback flows more freely, but one where attention reliably lands on the work and its standard. Build that, and you can afford a lot more feedback. Skip it, and every additional channel, human or machine, adds volume to a system that is already backfiring a third of the time.

Want to see how your leadership team could learn this on its own live work rather than in a seminar? Schedule a short demo and we will show you how Management Kits supports exactly this kind of working-and-learning loop.

Reference: 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.