The Feedback Sandwich, Reviewed: Where It Came From, Why It Spread and What the Evidence Found

What's the most widely used feedback technique in management? By a distance, the sandwich: say something nice, deliver the criticism, close with something nice. It has been taught on supervision courses for forty years, and most people reading this will have been on one end of it in the last month.

It also has a distinction few management techniques share. The feedback sandwich sits directly in the path of one of the most sobering findings in organisational psychology – a meta-analysis showing that more than a third of feedback interventions made performance worse – and the mechanism that study identified runs straight through the sandwich's middle. This review is written, as ever, generously: the technique was invented by kind people solving a real problem. But the evidence deserves to be laid out in full, because this may be the clearest case in this series of a tool that spread on comfort rather than results.

Where it came from

What is the feedback sandwich method?

The feedback sandwich wraps criticism between two layers of praise: say something nice, deliver the correction, close with something nice. Its most traceable champion is Mary Kay Ash, whose 1984 book advised managers to sandwich criticism between compliments. It spread because it solves the giver's problem - criticism is uncomfortable to deliver, and the sandwich makes the deliverer feel kind while doing it. Its benefits to the receiver were assumed rather than tested.

The sandwich has no technical report and no validation study, because it was never an instrument – it was folk wisdom with a memorable shape. Its most traceable champion is Mary Kay Ash, founder of Mary Kay Cosmetics, whose 1984 book Mary Kay on People Management advised managers to sandwich criticism between two layers of praise. The advice travelled because Ash was one of the most successful businesspeople of her era and because the technique solved the manager's problem beautifully: criticism is uncomfortable to deliver, and the sandwich makes the deliverer feel kind while doing it.

That sentence contains the review in miniature, so it's worth reading twice. The sandwich was designed around the giver's discomfort, and its benefits to the receiver were assumed rather than tested. For a technique that would be taught to millions, the assumption deserved a look. In 1996, it got one – though not aimed at the sandwich specifically, and all the more damning for that.

The study that changed the question

Does the feedback sandwich actually work?

The evidence says no. The largest meta-analysis of feedback ever run - Kluger and DeNisi, 607 effect sizes across 23,663 observations - found over 38 per cent of feedback interventions made performance worse, and identified praise as a moderator that weakens feedback because it pulls attention to the self rather than the task. Direct tests found the sandwich's ordering underperforms and its praise is quickly decoded as a warning that criticism is coming.

Avraham Kluger and Angelo DeNisi's 1996 paper in Psychological Bulletin, "The Effects of Feedback Interventions on Performance", remains the largest examination of feedback ever conducted: a review of nearly a century of research, distilled into 607 effect sizes from 131 studies covering 12,652 participants and 23,663 observations.

The average finding was positive – feedback improved performance by a moderate 0.41 standard deviations overall. Then comes the sentence that should be printed on the wall of any management training room: over 38 per cent of the feedback effects were negative. In more than a third of cases, giving people feedback on their performance made their subsequent performance worse than giving them none at all. Feedback, in other words, resembles a medicine with a strong average benefit and a very high rate of adverse reactions – and for decades the profession had been prescribing it without reading the label.

Kluger and DeNisi went further than counting; they proposed a mechanism, which they called Feedback Intervention Theory. Attention during work operates at three levels: task-learning (the details of doing the work), task-motivation (effort and persistence on the work) and meta-task (the self – what this says about me). Their central finding: feedback helps when it directs attention down the hierarchy, towards the task, and harms when it directs attention up, towards the self. When feedback becomes information about me rather than about the work, cognitive resources drain away from the task and into managing the threat – rumination, face-saving, discounting the messenger.

And here is the finding that lands directly on the sandwich: praise was one of the moderators that weakened feedback's effects, precisely because praise directs attention to the self. So the sandwich's defining move – wrapping task information in two layers of self-directed comfort – is, on the largest body of evidence available, aimed in exactly the wrong direction. The bread pushes attention towards the self; the filling needed it on the task.

What the direct tests found

Why does the feedback sandwich fail?

Two mechanisms, both well documented. First, attention: feedback helps when it keeps attention on the task and harms when it pulls attention to the self - and praise is a self-directed signal, so the bread aims the conversation in exactly the wrong direction. Second, decoding: with repetition, people learn that a compliment from their manager is the first slice of bad news, so the praise stops reassuring and the criticism arrives softened to the point of vagueness.

Kluger and DeNisi tested feedback in general. The smaller literature that examined the sandwich itself points, with one interesting exception, the same way.

Von Bergen and colleagues (2014), in a paper titled "The Sandwich Feedback Method: Not Very Tasty", assembled the case against: the praise is quickly decoded as a warning of incoming criticism (managers who use the technique train their teams to flinch at compliments), the criticism is softened to the point where it sometimes fails to arrive at all, and the closing praise blurs what was said in the middle. Studies of message ordering back this up. Henley and DiGennaro Reed (2015) tested the sequences experimentally and found the classic positive–corrective–positive ordering underperformed: the arrangement that worked best put the corrective information first, followed by the positives. Primacy and recency effects – the tendency to remember beginnings and endings – mean the sandwich buries its one piece of useful information in the position people remember least.

The exception deserves its paragraph, because reviews in this series report the evidence both ways. Prochazka and colleagues (2020) found that students given sandwich-style feedback prepared more thoroughly for a subsequent task than those given corrective feedback alone or none. One reasonable reading: when the recipient is a stranger and the relationship offers no other source of goodwill, the bread may buy enough motivation to outweigh its costs. That's a real result, and it hints at why the technique feels like it works. The catch is that the workplace case for the sandwich rests on repetition – and repetition is exactly what decodes it. The tenth sandwich from the same manager carries no reassurance at all; it carries a signal that criticism is coming, wrapped in phrases the receiver has learned to wait out.

Practitioner opinion has largely caught up. Kim Scott's Radical Candor argues against the technique on the grounds that it undermines trust in both the praise and the criticism; behavioural psychologist Aubrey Daniels made the same case a decade earlier. When the people who write the feedback books and the people who run the meta-analyses agree, the question stops being whether to retire a technique and becomes what to put in its place.

Why it survives anyway

It's worth pausing on this, because the sandwich's persistence is the most interesting part of its story. Three reasons, in my experience. It manages the giver's anxiety, and the giver chooses the technique. It produces a pleasant-feeling conversation, and pleasant-feeling conversations tend to get rated as successful on both sides of the desk – Kluger and DeNisi's data measured performance afterwards, which is where the damage shows. And it contains a true insight, poorly implemented: relationships do need more positive than negative in them, and people do hear hard messages better from someone who plainly values them. The error was converting that truth about relationships over time into a formula for single conversations. The repair, fortunately, follows directly from the same research.

What to do instead - cited

What should managers use instead of the feedback sandwich?

Five evidence-backed practices: aim every sentence at the task and keep the person out of it; use a structure that enforces this, such as Situation-Behaviour-Impact; if correction and praise must share a conversation, put the correction first, where attention is freshest; deliver praise separately, tied to specific work, and mean it; and close by handing the task back with a question - what would you change first? - rather than a cushion.

The evidence points to five practices. This is the section to copy.

1. Aim every sentence at the task, and keep the person out of it. The single strongest lever in Kluger and DeNisi's data: feedback that holds attention at task level helps; feedback that pulls attention to the self harms. "This report needs the numbers checked" beats "you're careless with detail" in every way the research can measure.

2. Use a structure that enforces that aim. The Center for Creative Leadership's Situation–Behaviour–Impact model does the job: name the specific situation, describe the observable behaviour, state its impact. No judgements on character, no bread.

3. If you must combine correction and praise, put the correction first. Henley and DiGennaro Reed's ordering experiment found corrective-first sequences most effective and praise-first least. The receiver's attention is freshest at the start; spend it on the message that matters.

4. Deliver praise separately, and mean it. Praise is not the villain here – instrumental praise is. Recognition given on its own day, tied to a specific piece of work, builds exactly the relationship capital the sandwich tries to counterfeit. The positivity belongs in the relationship, not in the delivery mechanism.

5. Close the loop with a question, not a cushion. Feedback that ends by handing the task back – "what would you change first?" – keeps attention where the evidence wants it: on the work, and on what happens next. (Readers of our GROW review will recognise the shape.)

What to take from it

Where this lands, then. The feedback sandwich deserves a kinder obituary than it usually gets: it was built by people who understood that criticism lands badly in cold relationships, and that insight survives its messenger. As a technique, though, the record reads one way. The largest meta-analysis in the field found that feedback fails more than a third of the time, identified attention-to-self as the mechanism, and flagged praise as a moderator that pushes attention in precisely that direction. The direct tests found the format's ordering backwards and its praise quickly decoded. The one supportive study speaks to first encounters rather than the repeated relationships management is made of.

Retire the formula. Keep the kindness, and move it where it works: into specific, task-level feedback delivered plainly, and into praise generous enough – and separate enough – that no one ever hears it as the first slice of bad news.

Six questions before you give any feedback

Copy, paste and run through these before your next difficult conversation:

1. Is every sentence I plan to say about the work, rather than the person?
2. Can I name the specific situation and behaviour – or do I only have an impression?
3. Have I put the important message first, rather than cushioned in the middle?
4. Would this praise survive on its own, on a different day? (If it only exists to soften what follows, cut it.)
5. Does the conversation end with the task handed back – a question, a next step, a date?
6. When did this person last hear specific praise from me that had no second act?

Where our thinking sits

Readers of this site will know where Sariio stands: feedback lands differently on different people, and the evidence above is about averages. A current reading of how each person prefers to receive information – directly or gently, in the moment or with time to prepare – is the piece these formulas were reaching for. That argument is made across our other essays; the research in this one stands on its own.