Giving Peer Feedback That Actually Helps

"You should be a bit more proactive." Six words, said at the end of a meeting, with good intentions. The colleague nods. Three months later nothing has moved and both people are slightly more awkward around each other than before.

That feedback failed for precise reasons. It says nothing about when, about what, or about who has a problem with the situation. It judges a character trait and leaves the colleague guessing what to do differently on Monday morning.

Peer feedback gets practised in just about every team that calls itself horizontal, with wildly uneven results. Research has documented that gap for thirty years, and it gives a fairly clear picture of what separates useful feedback from feedback that damages the relationship.

Feedback does not improve performance by default

Avram Kluger and Angelo DeNisi published the meta-analysis that still anchors the subject in Psychological Bulletin in 1996. Across 607 effect sizes and 23,663 observations, feedback interventions improve performance on average, with a d of 0.41. In more than a third of cases, they make it worse.

Their explanation rests on a simple distinction. Feedback that pulls attention back to the task helps the person correct their work. Feedback that pulls attention back to the self triggers a defence of self-image, and the energy goes there instead of into the work. "You should be more proactive" aims at the self. "The client chased us about the quote before we replied" aims at the task.

360-degree feedback went through the same scrutiny. James Smither, Manuel London and Richard Reilly reviewed the available data in Personnel Psychology in 2005 and concluded that improvement in ratings over time is generally small.

A third finding complicates matters further. Steven Scullen, Michael Mount and Maynard Goff showed in the Journal of Applied Psychology in 2000 that the largest share of variance in performance ratings comes from who is rating, more than from the person being rated. That is the foundation of the critique Marcus Buckingham and Ashley Goodall made in "The Feedback Fallacy" (Harvard Business Review, March-April 2019). Your feedback says as much about you as about your colleague.

None of this condemns the practice. It sets the bar. Useful feedback describes a piece of work, it can be checked, and it ends with something the colleague can do.

Two conditions to meet before you speak

Your colleague can contradict you without risk

Amy Edmondson defined psychological safety in Administrative Science Quarterly in 1999 as the shared belief that people will be neither punished nor humiliated for raising an idea, a question, a concern or a mistake. When that belief is missing, the colleague nods in the meeting and carries on exactly as before, because arguing costs more than staying quiet.

Gallup has tracked a neighbouring item in its US surveys for years. Only three employees in ten strongly agree with the statement "at work, my opinions seem to count". The article on the structural problems of hierarchy covers what organised silence costs a company and where it comes from.

You share a reference for what was expected

Feedback on a piece of work assumes a prior agreement on what that work was meant to produce. When the expectation is written nowhere, "you should have warned the team" becomes one personal preference facing another personal preference, and the conversation ends in a draw.

That is the job of accountabilities, the ongoing activities others expect from a role. "Answering press requests within 48 hours" can be checked by opening the inbox. "Being responsive" can be argued about forever. The article on roles and job descriptions explains where the line between those two documents runs.

Wording that holds up

The Center for Creative Leadership has long taught the SBI model, for situation, behaviour, impact. You place the moment, you describe what you saw or heard, you state the effect it produced. The SBII variant adds a fourth step, intent: instead of assuming it, you ask for it, with a question along the lines of "what were you trying to do at that point?".

The gap between vague feedback and usable feedback shows up better on examples.

What was said	What is missing	What you can say instead
"You should be more proactive."	A date, a fact, an effect. The rest judges a character trait.	"On Tuesday the client chased us about the quote before we replied. I improvised an excuse on the phone. What got in the way on your side?"
"Your specs are always sloppy."	"Always" closes the discussion and the adjective aims at the person.	"On the payment module spec, the error cases were missing. We found them during QA and shipped two days late."
"You don't communicate enough."	The expectation stays inside your head.	"I hear about schedule changes in the weekly meeting, sometimes after the client does. I would like to read them in the role thread when you decide them."
"Great job, keep it up!"	Praise with no content cannot be reused.	"Your review of the contract surfaced the automatic renewal clause. We avoided a year of commitment. Do exactly that again on the two September contracts."
"I think you lack rigour."	A global judgement, unverifiable and impossible to correct.	"The March report gave three figures that differed from the source spreadsheet. I spent an hour finding the right one."

Three habits make the difference in those rewrites. Speak in the first person, because "I spent an hour" is far harder to dispute than "it is unreadable". Give the date, because a dated fact can be checked and a recent dated fact can still be corrected. Ask about the intent before you invent it, because half the feedback conversations that go wrong rest on an intent wrongly attributed.

Timing counts as much as wording. Feedback given within a few days covers a situation both people still have in mind. Feedback saved for the annual review lands on a situation nobody can reconstruct.

This video walks through a feedback conversation, step by step.

What the sandwich and the positivity ratios are worth

Everyone knows the sandwich recipe: a compliment, the criticism, a compliment. Its empirical base rests on very little. Jakub Prochazka, Martin Ovcari and Michal Durinik tested it in 2020 in Learning and Motivation, on 91 students solving maths problems with written feedback delivered by computer. The group that got the sandwich prepared longer for the second set and solved more problems than the group that got the corrective part alone. The authors themselves call that evidence partial and ask for replications, because the experiment tests a single condition and says nothing about whether the effect comes from the order of the pieces or from the mere presence of positive sentences. A habit backed by one laboratory experiment is still a habit.

The figure of 5.6 compliments per criticism has circulated since a piece by Jack Zenger and Joseph Folkman in the Harvard Business Review in March 2013, "The Ideal Praise-to-Criticism Ratio". The family of numbers it belongs to has aged badly. Barbara Fredrickson and Marcial Losada had published a critical ratio of 2.9013 in American Psychologist in 2005, supposedly separating people who flourish from people who stagnate. Nick Brown, Alan Sokal and Harris Friedman took the demonstration apart in the same journal in 2013, and American Psychologist published a note that year declaring the ratio and its upper bound invalid.

So no compliment quota is waiting for you. What makes praise useful is that it names a precise act the colleague can repeat.

When feedback lands nowhere

Well-worded feedback can still produce nothing. Four situations hide behind that silence, and they call for different follow-ups.

• Your colleague remembers something else. Go back to the written trace, the meeting minutes, the discussion thread, the delivery date. When no trace exists, the conversation stops there and you know what to put in place for next time.

• You thought warning the team was part of the role, your colleague thought the opposite. The subject has changed nature: it is an accountability to write down or to remove. Take it to a governance meeting instead of repeating it one to one.

• Your colleague heard you and has other priorities. Ask which ones, and by what date they plan to come back to the subject. A dated answer beats a nod.

• The same gap keeps coming back and costs several people. Step out of the one-to-one and raise it as a tension in the role concerned, where others can process it.

In the first three cases, close with the same question. What do we do, and by when? Feedback with no next step evaporates within a week.

Receiving feedback is a practice too

The person receiving decides what the feedback becomes. Rephrase before you answer, in your own words, to check you are both talking about the same thing. Ask for a dated example when the feedback arrives as an adjective. Then say what you are going to do, or say that you will do nothing and why.

That last answer is perfectly legitimate and far more useful than a polite nod. A colleague who says "I am keeping this priority, I will give earlier notice on client deliveries and keep internal tasks as they are" has just made what happens next predictable for everyone.

What feedback cannot fix

Some subjects arrive disguised as feedback between colleagues and leave untouched.

An unmanageable workload belongs to how roles are distributed and gets handled by revising who carries what. A disagreement about pay or job level belongs to the employment relationship and gets handled with the person who decides. An entrenched interpersonal conflict needs a third party, and one more conversation between the same two people usually makes it worse.

Putting those subjects where they belong is what keeps feedback credible as a practice in a team.

What makes feedback discussable

Feedback gets discussed calmly when both people are reading the same document. That is the only place where a tool changes anything.

In Rolebase, a role carries its purpose, its domain and its accountabilities, and the org chart says who holds it today. Feedback anchored to a written accountability covers a shared expectation instead of an impression. Meetings follow steps and open with a round where everyone speaks before the discussion starts, which gives a regular slot to the subjects that drag on. Decisions and threads stay readable, so finding what was agreed and on what date takes a few seconds.

Where to start

1. Write the accountabilities of the two or three roles where friction comes back most often. Feedback anchored to a written expectation gets discussed, feedback anchored to an implicit one gets argued over.
2. Take a piece of feedback you have been chewing on for two weeks and write it as situation, behaviour, effect. Three sentences are enough.
3. Give it within a few days of the next occurrence, and ask about the intent before assuming it.
4. Close with a dated next step and write it where you will find it again, in the role thread or in the meeting minutes.

Frequently asked questions

Does peer feedback really improve performance?

Not automatically. The meta-analysis by Avram Kluger and Angelo DeNisi, published in Psychological Bulletin in 1996 across 607 effect sizes and 23,663 observations, finds a positive average effect (d = 0.41) and records that over a third of feedback interventions made performance worse. Their theory explains the gap by what the feedback aims at: feedback that directs attention to the task helps, feedback that directs it to the person triggers a defence of self-image.

Should you use the sandwich method?

Nothing requires it. A single published experiment tests it directly, the one by Jakub Prochazka, Martin Ovcari and Michal Durinik in Learning and Motivation in 2020, on 91 students receiving computer-delivered written feedback on a maths test. The sandwich group did better than the group that got only the corrective part, and the authors call that evidence partial while asking for replications. Feedback that is clear, dated and followed by a next step produces better results than the wrapping.

How many positives for one criticism?

No ratio has any authority. The 5.6 to 1 figure comes from a piece by Jack Zenger and Joseph Folkman in the Harvard Business Review of March 2013. The most cited ratio in that family, the 2.9013 published by Barbara Fredrickson and Marcial Losada in American Psychologist in 2005, was declared invalid by that same journal in 2013, after the demonstration by Nick Brown, Alan Sokal and Harris Friedman. Aim for praise that names a precise act rather than for a quota.

How often should you give feedback?

Gallup measures that 80% of employees who received meaningful feedback in the past week describe themselves as fully engaged, and that employees whose manager gives daily rather than annual feedback are 3.6 times more likely to say they are motivated to do outstanding work. Those figures cover the relationship with the manager, rather than exchanges between peers. The practical rule that follows still holds between colleagues: talk while the situation is still fresh for both of you.

What do you do when a colleague rejects the feedback?

Start by finding what the disagreement is about. On the facts, go back to the written trace. On the expectation, the subject becomes an accountability to write down or to remove, and it gets handled in a meeting rather than one to one. On priorities, ask for a date. When the same gap keeps returning and affects several people, raise it as a tension in the role concerned instead of repeating the same conversation.

How does this differ from the OSCAR method?

OSCAR is a named five-step interview frame, described in our article on feedback with the OSCAR method. It structures a conversation from beginning to end. The SBI model presented here structures a single contribution, the one where you lay out a fact and its effect. The two combine without difficulty: SBI supplies the sentences, OSCAR supplies the sequence.

What matters

Peer feedback works when it covers a described piece of work, on a known date, with a named effect and an agreed follow-up. It fails when it covers a character trait, when nobody had agreed on what was expected, and when the conversation ends on a nod.

The most thankless part of the work happens before the conversation. Writing the roles and their accountabilities gives both people the same starting point, and that is exactly what Rolebase makes concrete: a living org chart where every role carries its purpose, its domain and its expectations. Create your organisation free for up to five active members.