Feedback model cheat sheet: SBI, COIN, STAR, DESC, AID

Most managers end up collecting these rather than using them. You pick up SBI in one training, COIN in a book, DESC in a workshop, and they blur into a general sense that feedback is supposed to have steps — so you use none of them, and improvise instead. The fix isn't memorising all five. It's noticing that they differ in exactly one place: where they stop. Every one of them makes you describe a specific behaviour rather than judge a person, which is the part they share and the part that matters most. What separates them is the ending, and the ending is chosen by the situation, not by taste.

The cheat sheet

The one question that picks for you

1. 1

   Do you already know what should change?

   If no, stop at SBI. State the observation, then be quiet and ask — the person doing the work usually has the better fix, and prescribing before you've heard it is how managers solve problems nobody had.
2. 2

   Is it yours to prescribe?

   If you're their manager and the fix is procedural, AID is the fastest thing here. Sideways to a peer or upward to your own boss, stating a desired outcome reads as instruction, so make it a request instead: that's DESC.
3. 3

   Does the plan need their fingerprints on it?

   Recurring problems and development conversations don't survive on your plan alone. COIN's fourth step exists to make the ending something you both signed, which is what makes it hold past the meeting.
4. 4

   Are you actually recognising something?

   Then none of the corrective models fit. STAR is the one built to make praise specific enough to repeat — and specific praise is the highest-return, lowest-cost thing on this page.

The same moment, five endings

One situation, so the difference is visible rather than theoretical. A report ran the client workshop, cut the prioritisation exercise for time, and the client left without agreeing what ships first.

SBI — "In Thursday's workshop the prioritisation exercise got cut for time, and the client left without agreeing what ships first." Then nothing. The silence is the tool.

AID — "…and the client left without agreeing what ships first. Next time protect that slot, even if it means dropping the warm-up."

COIN — "…left without agreeing what ships first. What would make room for it next time?" The ending is whatever you both land on.

DESC — "…so I'm now chasing the priority call over email. Could you keep the prioritisation slot fixed and cut the warm-up instead? Then I don't have to reopen it with them."

STAR/AR — "You had ninety minutes and a client who arrived twenty late, you kept discovery whole and cut prioritisation, and we left without a decision. Cutting the warm-up instead would have bought you both."

Same facts in all five. The one you want depends entirely on what has to be true when the conversation stops.

Direction changes the answer more than the content does

This is the cut most comparisons skip, and it's the one that decides whether a model helps or backfires.

When none of these is the right tool

A cheat sheet is only honest if it says where it stops. Three moments where reaching for any of the five makes things worse:

It's actually a warning. If the real message is that something has to change or there will be consequences for someone's job, say that plainly. Dressing it as a communication model is unkind, because the person leaves the room having heard a technique instead of the stakes. Feedback down to a report runs on restraint precisely so the heavier conversation stays legible when it arrives.

The cause isn't awareness. All five models assume the person didn't know, or didn't see the impact. When the cause is capability, unclear priorities, or a constraint you set yourself, a better-worded note changes nothing — and the third repetition of well-structured feedback fails the same way the first two did.

You're describing a pattern, not a moment. Every model here is built around one specific instance. Three instances across different contexts is a different conversation, and it needs naming as a pattern rather than a fourth well-anchored example.

What all five are actually for

Underneath the acronyms there's one shared mechanic, and the evidence for it is older than any of these frameworks. The classic meta-analysis by Kluger and DeNisi found that feedback improves performance on average — but made it worse in over a third of cases, and the dividing line was where the feedback pointed: at the task and the behaviour it helped, at the person it hurt. Every model on this page is a device for keeping your attention on the first thing when your instinct wants the second.

Which also tells you what they can't do. A model makes you specific, and specificity is the easy half. Whether the person believes you're on their side is the hard half, and no acronym supplies it — that's a different diagnosis entirely, and it's the one to run first when well-structured feedback keeps failing. If what you want is the wording, a dozen worked rewrites show the swap running on real moments. And if you want suggestions about the next attempt rather than an account of the last one, that's not on this page at all — it's feedforward, a different act with different rules.

Where Mirorly fits

Read the table again and notice who's missing from it. All five models are tools for the person talking. Not one of them describes what happens when the feedback needs to travel toward you — and the direction column explains why: upward is the hardest case, so the observations about how you work are the ones least likely to be spoken out loud. You can get fluent in all five and still be the least-informed person about your own behaviour.

Mirorly is built for that direction. You answer specific behavioural questions about how you work, the people around you answer the same set, responses are aggregated with an anonymity threshold so honesty carries no cost, and what you read is the gap between your answers and theirs. Same discipline as the models — anchored to behaviour, not character — pointed the way none of them go. The core leadership behaviours template is where most people start.

Common questions

The one-line summary

Five models, one distinction: SBI ends on an observation and hands over the silence, AID ends on your instruction, COIN ends on an agreement you both made, DESC ends on a request plus what improves if you get it, and STAR ends on a complete story for recognition or a decision-point lesson — so choose by the ending the moment needs, check it against direction (AID downward only, DESC sideways and upward, SBI anywhere), and remember that all five do the same underlying job of keeping your attention on behaviour rather than character, which is the half of feedback that a framework can actually fix.