The feedback sandwich is one of the most durable pieces of bad management advice ever written down. Start with a compliment. Deliver the criticism. Close with another compliment. Serve the hard thing between two slices of nice.
Almost every manager has been taught it. Almost no one likes receiving it. And when researchers actually measure what it does, the results are worse than just delivering the criticism directly.
Why the Sandwich Fails
Research in Organizational Behavior Management has shown for over a decade that the sandwich method weakens, not strengthens, the performance effects of feedback. A 2025 review summarized the mechanism plainly: most employees know the positive feedback is a buffer, and the compliments start to feel like empty appetizers before the inevitable "but" (Phys.org, 2025). They end up ignoring the positive and resenting the negative.
There are three specific problems the research consistently identifies:
The structure is recognizable. Once someone has been on the receiving end of two or three sandwiches, they pattern-match it instantly. "You have been doing great work on X" becomes a flag that bad news is coming, which poisons future praise. The ABA Technologies analysis of the sandwich put it bluntly: the method can train employees to distrust praise entirely (ABA Technologies).
The critical message gets diluted. Sandwiching makes it easier for the recipient to walk away unclear about what actually needs to change. Researchers have found that the method often gives employees a diluted understanding of their work performance (The Sandwich Feedback Method: Not Very Tasty).
It protects the giver, not the receiver. The sandwich is a comfort object for the manager. It lets you feel like you softened the blow without actually improving the outcome. Kim Scott's team at Radical Candor noted the same pattern: the sandwich is about your discomfort giving the feedback, not about the other person's ability to act on it (Radical Candor).
None of this means you should stop giving praise. It means praise and criticism should live in separate conversations, each delivered for what it actually is.
The Three Frameworks That Do Work
Three frameworks have replaced the sandwich in research and in the field. They are not competitors. They operate at different layers of the feedback conversation.
SBI: Situation-Behavior-Impact
SBI was developed by the Center for Creative Leadership and has become the default in-moment framework for delivering specific behavioral feedback (CCL). It has three parts.
Situation: anchor the feedback to a specific moment. "In yesterday's design review" is better than "lately." Specificity reduces the ability to dismiss the feedback as vague.
Behavior: describe the observable action. "You interrupted Maria three times before she finished her proposal." Not "you were being aggressive." Behavior is what a camera would have captured. Everything else is interpretation.
Impact: connect the behavior to what happened as a result. "That signaled to the team that her ideas were not worth hearing in full, and the junior engineers went quiet for the rest of the meeting."
The reason SBI works is that it separates what happened from what you think about it. Most defensiveness is a response to feeling judged as a person. SBI keeps the conversation on actions and outcomes, which is harder to argue with.
COIN: Context-Observation-Impact-Next
COIN is SBI with a fourth step. Context replaces Situation, Observation replaces Behavior, Impact stays, and a Next step is added at the end.
The "Next" step is the meaningful addition. It converts the feedback from diagnosis to forward motion. "For the next design review, what would it look like if you let each person finish their proposal before responding?" is a different conversation than "please don't interrupt people."
COIN works best when the feedback is about a recurring pattern and the goal is a visible behavior change. SBI alone can feel like it ends without resolution. COIN forces the conversation toward what comes next, which is what the recipient is going to be asking themselves anyway.
Radical Candor: Care Personally + Challenge Directly
Kim Scott's Radical Candor is not an in-moment framework. It is the relational frame SBI and COIN assume but do not build (Radical Candor). The core idea is a 2x2 matrix: Care Personally on one axis, Challenge Directly on the other.
The four quadrants map how feedback lands:
Radical Candor (care + challenge): the feedback is both honest and grounded in a real relationship. The recipient can hear it because they trust the intent.
Obnoxious Aggression (challenge without care): the criticism is clear but feels like an attack. The recipient defends instead of processing.
Ruinous Empathy (care without challenge): the manager softens the message so much that the actual issue never lands. This is where the sandwich lives.
Manipulative Insincerity (neither): the manager says whatever keeps the peace in the moment. Eventually discovered, always resented.
Radical Candor matters because the same SBI script can land as insight or as attack depending on whether the relationship underneath it has been built. The 2x2 is a reminder that feedback quality is a function of both content and trust.
Comparing the Frameworks
| Framework | When to Use | Strength | Weakness |
|---|---|---|---|
| Sandwich Method | Rarely. Only when the relationship is very new and the stakes are very low. | Feels safe for the giver. | Recognizable pattern, dilutes the message, trains distrust of praise. |
| SBI | One-off moments. A specific incident you want to name clearly. | Specific, behavioral, low defensiveness. Works across seniority levels. | Can end without a clear path forward if you stop at Impact. |
| COIN | Recurring patterns. When you want the conversation to end with a concrete next step. | Forward motion. Closes the loop on behavior change. | Can feel prescriptive if the Next step is handed to the recipient instead of co-created. |
| Radical Candor | Ongoing manager-report relationships. The frame for how feedback gets delivered over time. | Names the relational conditions that make feedback land at all. | Not an in-moment script. Needs SBI or COIN inside it to deliver specific feedback. |
The pattern in practice: Radical Candor as the relational frame, SBI or COIN as the script for any given conversation, sandwich method left on the shelf.
When the Recipient Gets Defensive
Defensiveness usually signals that the person feels judged as a person rather than informed about an action. It rarely means the feedback itself was wrong. Three moves help when you see it.
Slow down and restate the behavior. "I am not saying you are dismissive. I am describing what happened in that meeting and what I saw as the effect." This pulls the conversation back to observable action.
Ask one genuine question. "Walk me through what was going on for you in that moment." Not as a gotcha. As a real question. Defensiveness usually softens when the other person feels heard. Most managers try to fix defensiveness by piling on more data, which makes it worse.
Name the gap between intent and impact. This is the reason the Center for Creative Leadership added the "Intent" extension to SBI (calling it SBII). "I believe your intent was to get to a decision faster. The impact in the room was that the junior engineers stopped contributing." Separating intent from impact lets the person acknowledge the impact without feeling like you are attacking their character.
A 2025 Peaceful Leaders Academy review of feedback research found that 92% of employees say constructive feedback improves their performance, but only when it is specific and actionable (Peaceful Leaders Academy). Defensiveness is almost always a sign the feedback has drifted into character judgment. The fix is to return to behavior.
Giving Criticism to Someone More Senior
The hardest criticism conversation is the one going up. Your skip level interrupted the engineer. Your VP took credit in a cross-functional meeting. Your founder rewrote the strategy document without talking to the team.
Three moves make it possible:
Ask permission before delivering. "I have an observation from yesterday's meeting that I think might be useful. Can I share it?" This is not a formality. It signals that you have thought about whether the moment is right and gives the senior person a chance to land somewhere they can actually hear you.
Anchor to a shared outcome. "You have said you want the engineers speaking up more in reviews" is a stronger opener than "you interrupted Maria." The shared goal puts you on the same side of the feedback.
Use the same SBI structure. Seniority calls for more specificity about the situation and more care about the impact framing, not softer observations. Most senior people would rather receive a direct, behaviorally grounded observation than vague hedging. The vague version reads as political caution, which is worse than directness.
The biggest failure mode is going sideways instead of up. If you are venting to peers about a senior person's behavior instead of delivering the feedback directly, you are contributing to the problem you are complaining about.
Where QuestWorks Fits
The hardest part of constructive criticism in practice is that the feedback comes from a human whose motives the recipient has to decode. "Is my manager saying this because it is true, or because they are annoyed?" That question sits underneath every piece of feedback and is the main reason defensiveness spikes.
QuestWorks, the flight simulator for team dynamics, was designed so a significant portion of behavioral feedback does not come from a human at all. The platform runs 25-minute scenario-based quests on its own cinematic, voice-controlled interface. Groups of two to five practice decisions, coordination, and conflict under pressure. After each quest, the debrief surfaces what happened: who communicated proactively, who escalated too fast, where coordination broke down.
Because the feedback comes from behavioral data rather than someone's impression, it lands differently. There is no ego in the middle. "Here is what happened when you tried that approach" is easier to act on than "I think you were too controlling in that meeting." The AI facilitator surfaces patterns the way game analytics surface player performance. The manager then has specific, recent behavioral reference points to use in 1:1s, which is the material SBI and COIN both depend on.
HeroGPT, the private AI coaching layer that integrates with Slack, is the complement. It never shares upstream. A manager who wants to rehearse how to give a hard piece of feedback before the real conversation can think out loud, try different framings, and get private coaching before they deliver it. Reps without stakes, then the stakes.
Participation is voluntary. Quests are not tied to performance reviews. QuestDash is strengths-based and visible to everyone including players. The feedback data that shows up is behavioral, specific, and owned by the person it describes, which is exactly what the research says feedback needs to be to actually change performance.
The Underlying Principle
Across every framework in this article, the same principle holds. Feedback that lands describes behavior, connects it to impact, and stays out of judgments about the person. The sandwich method fails because it is structured around the giver's comfort. The frameworks that work are structured around the receiver's ability to act on what they heard.
If you strip everything else away, constructive criticism is this: say what you saw, say what it led to, say what comes next. Do it in a relationship the other person trusts. Skip the bread.
Related: How to give feedback to engineers that actually lands.
QuestWorks: $20/user/month, 14-day free trial. Integrates with Slack.
Try it free
