Which feedback model should I use for a difficult conversation?

This article is a gateway into the wider set of approaches in my feedback models hub, where each model has its own guide.

Perhaps “best” is the wrong question

I have lost count of the leaders who can name six feedback models and still avoid the conversation. Part of the problem is that “which is best?” sounds like a question about the models, when it is really a question about the conversation. AID, CEDAR, BIFF, COIN and the rest share a good deal of underlying discipline: describe something specific, explain why it matters and work out what happens next. What differs is what each one is built to do with that discipline, and that is where the choice lies.

For that reason I treat them less as competing frameworks and more as a small toolkit, a difficult conversation framework you assemble from parts. Any constructive feedback model earns its place by fitting the job, and different conversations are difficult for different reasons. Sometimes you are unsure whether your evidence is sound. Sometimes you know what happened but not why. Sometimes you are braced for a strong reaction. Sometimes you would rather the other person reached the conclusion themselves. Naming which of these applies is usually more useful than asking which model is best.

Start with what is making the conversation difficult

The table below starts from the difficulty rather than the model. It is deliberately short. My hub has a fuller version covering the wider family, including E2C2, AIR, CORBS, FAST, CLEAR and DESC.

Prepare first, then choose

Preparation comes before choosing a delivery model. The BOOST feedback checklist asks whether what you plan to say is Balanced, Observed, Objective, Specific and Timely, and I think of it as the private test that comes before any of the delivery models. If AID is the script, BOOST is the check that the script is worth delivering. It is not designed to be spoken aloud, which is why it sits alongside the others rather than competing with them.

The most useful question it prompts is a simple one: can I describe a specific example of what I actually saw? If I cannot, I am probably not ready, whichever delivery model I choose later. The same applies to my own state. When I notice that I am angry or anxious, I would usually rather wait a day than deliver feedback shaped by the emotion, since objectivity is one of the first things to go.

When you know what happened: AID

If the issue is a specific, observable behaviour and the cause is not really in question, AID is where I usually begin. It asks for one Action you observed, its Impact, and the Development or desired behaviour that follows, ideally suggested by the other person before you supply it. It is quick enough to use in the flow of normal work, and because it stays with what happened rather than what you think it says about the person, it is hard to argue with.

Its limit is that it assumes your picture is reasonably complete. If the response to your first sentence is “that’s not what happened” or “there’s more to it than that”, you have probably learned something important about how much you know. That is often the moment to move towards CEDAR rather than to repeat the action more firmly. E2C2 and AIR use different language for very similar territory, and you will find both, with the rest of the wider family, on the hub.

AID vs CEDAR: telling or finding out

The comparison I am asked about most is AID vs CEDAR, and I think the difference comes down to what you assume about your own picture of the situation. AID works well when you have enough of that picture to describe what happened and why it matters. CEDAR assumes you do not, or that the other person’s account is part of the answer. Its distinctive Diagnosis stage asks both of you to explore what happened and why before agreeing on actions, and the review, typically two to three weeks later, keeps the conversation going after the meeting ends.

A few signals suggest that CEDAR may be the better fit. You are looking at a pattern rather than an incident. The same conversation has happened before with little change. The person agrees readily but nothing shifts, which can be compliance rather than agreement. Or you suspect something underneath the behaviour that neither of you has yet named. The cost is time and effort. CEDAR is too heavy for a small correction, and remote conversations need extra care in the diagnosis stage because there are fewer cues to read.

One caution. The diagnosis has to be real. If you have already decided what the cause is, using CEDAR as a route to your conclusion tends to feel manipulative, and AID, used openly, is the better choice. My general rule is to use the lightest model that fits and to move up when what I hear tells me my picture is incomplete.

When you expect a strong reaction: BIFF

If I expect the conversation to be emotionally charged, I look for a model that builds in a pause. BIFF, as I use it on the hub, stands for Behaviour, Impact, Future expectation and Feelings. It describes what happened, why it matters and what should happen next, then deliberately stops to notice how the other person is receiving it before you carry on. That pause does not turn the conversation into therapy. It gives you a way to notice whether the person can still hear and process what you are saying, and to adjust your pace if they cannot.

A note on naming. The same letters are better known for Bill Eddy’s BIFF response, which stands for Brief, Informative, Friendly and Firm and is designed for replying to hostile emails, texts and posts (High Conflict Institute, n.d.). That is a different tool for a different job. The version I describe here is closest in spirit to John Tomlinson’s BIF framework (Tomlinson, 2014), with an explicit emotional check added. I would not reach for it for routine, low stakes feedback, where the pause can feel oddly formal.

When you want them to do the thinking: COIN

COIN moves through Context, Observation, Impact and Next, and some versions distinguish between a push and a pull approach (MindTools Content Team, 2025). In a push delivery you state your view directly. In a pull delivery you invite the person to reach the observation with you, which suits someone capable who probably already has a sense of what went wrong. I tend to choose it when I want the conversation to feel more like coaching than correction.

The risk with pull is being unclear about who holds the answer. If you already know exactly what you want the person to conclude, open questions can feel like a quiz with a hidden mark scheme. I would choose pull when I am prepared to accept what comes back, and push when the issue is clear enough that being direct is the more respectful option.

If you are the one receiving: RADAR

A difficult conversation is difficult from both sides, and the question of which model to use is easy to ask only from the giver’s chair. RADAR is designed for the person receiving feedback: Repeat what you heard, Ask clarifying questions, Discuss your own perspective, Align on what will change, and Reflect afterwards. Its value is that it slows the instinctive defensive reaction long enough to extract something useful. If you know you are about to be on the receiving end, it is worth rehearsing before the meeting, and my guide to receiving feedback with RADAR explores it in more depth. My Leadership Library page on responding to feedback covers the wider capability.

The relationship changes the choice too

The models do not change with the relationship, but what you emphasise does. When you are worried about defensiveness, the priority is separating observation from interpretation and leaving room for the other person’s view, which I explore in my article on feedback without defensiveness. When the person is someone you don’t manage, standing becomes the first question, and BOOST and AID help you stay with what happens where your work meets. With someone more senior, timing, channel and framing feedback as information matter as much as the structure. In each case the model gives you scaffolding, and the relationship determines how much weight it can carry.

What does current research suggest?

I have not found research that compares these named models head to head, which is not the same as saying none exists. Most seem to have grown out of practitioner experience, and I would be cautious about anyone claiming strong evidence that one mnemonic outperforms another. What the wider literature does offer is a few useful cautions about how to think about the choice.

Complexity is a recurring theme

A 25 year review by Anseel and Sherf (2025), subtitled “from simple rules to complex realities”, found that feedback research consists of several streams built on different assumptions, and that findings do not always map neatly onto how feedback is experienced in organisations. That seems to me a good reason to treat any model as a prompt for thinking rather than a formula.

What the models share may matter more than the letters

Kluger and DeNisi’s (1996) meta-analysis found that feedback improved performance on average, but that more than a third of the interventions reduced it. Their Feedback Intervention Theory offers an explanation: feedback becomes less effective as it shifts attention away from the task and towards the self. Every model discussed here is built to keep the conversation on behaviour and its effect, and that common feature may matter more than which letters you choose.

The harder problem may be having the conversation at all

Beck and Dahling (2025) argue that providing feedback that is useful and sensitive is a complex, motivated choice, and that it competes with other managerial demands, so it tends to receive less priority when pressure rises. I read this as one practical reason to value a model that is quick to recall, since it may lower the cost of starting. That is my interpretation rather than something the authors test.

Receptivity is shared

Fulham, Krueger and Cohen (2022) distinguish between recipients being receptive to feedback and being able to discern whether it is true and useful, and they identify barriers arising from the feedback itself, the person giving it and the person receiving it. The choice of model can reduce avoidable barriers on the giver’s side, particularly where it makes your reasoning visible or invites the other person’s account. It cannot guarantee that the recipient will be receptive.

Five questions before you choose a model

Before deciding how to structure the conversation, I would spend a few minutes on these five questions.

1. What do I want from this conversation? Awareness, understanding, a change, a commitment or a better relationship? The model should serve the answer rather than the other way round.
2. How much do I actually know? Separate what you observed from what you have inferred, and be open with yourself about what you do not yet know. The less you know about why, the more room you need for diagnosis.
3. How is this person likely to react? Consider what the feedback may threaten for them, and whether you need a model that builds in a pause.
4. How much time and standing do I have? A quick AID conversation and a CEDAR conversation with a review date are different commitments, and your relationship to the person affects how much weight either can carry.
5. Have I checked my own evidence and state? Run BOOST privately before you choose, and wait if emotion is shaping what you plan to say.

There is then one final practical decision. If I am unsure, I would usually prepare with BOOST, open with AID, and stay ready to move towards CEDAR if what I hear tells me the picture is incomplete.

Conclusion

The most useful answer to which feedback model to use is rarely a single set of letters. It is a short sequence: prepare so that your evidence and your state are sound, choose the lightest model that matches what the conversation needs, and stay ready to change approach when what you hear changes your picture. AID and CEDAR cover a great deal of ground between them. BIFF adds a pause when emotion is likely, COIN adds a choice about who does the thinking, and RADAR gives you something to hold on to when you are the one receiving.

None of them will make a difficult conversation comfortable, and I would be wary of any that claimed to. What they can do is reduce the avoidable difficulty, so that what remains is the difficulty that belongs to the subject itself.

Three questions to reflect on

1. What is making your next difficult conversation difficult: your evidence, your understanding of the cause, the likely reaction or your own state?
2. Which model do you reach for by default, and what would it look like to choose one on purpose?
3. When did you last change approach part way through a conversation because of what you heard, and what prompted it?

If you would like to compare the wider set of approaches, my feedback models hub brings them together, with fuller guides to AID, BOOST, CEDAR and RADAR. I also explore giving feedback in my Leadership Library resources.

Sources

• Anseel, F. and Sherf, E.N. (2025) ‘A 25-year review of research on feedback in organizations: From simple rules to complex realities’, Annual Review of Organizational Psychology and Organizational Behavior, 12, pp. 19–43. Available at: https://doi.org/10.1146/annurev-orgpsych-110622-031927 (Accessed: 21 September 2026).
• Beck, J.W. and Dahling, J.J. (2025) ‘Delivering high-quality feedback is a choice: A self-regulatory framework for understanding feedback provision in organizations’, Human Resource Management Review, 36(1), 101117. Available at: https://doi.org/10.1016/j.hrmr.2025.101117 (Accessed: 21 September 2026).
• Fulham, N.M., Krueger, K.L. and Cohen, T.R. (2022) ‘Honest feedback: Barriers to receptivity and discerning the truth in feedback’, Current Opinion in Psychology, 46, 101405. Available at: https://doi.org/10.1016/j.copsyc.2022.101405 (Accessed: 21 September 2026).
• High Conflict Institute (n.d.) ‘How To Write A BIFF Response®’. Available at: https://www.highconflictinstitute.com/how-to-write-a-biff-response/ (Accessed: 21 September 2026).
• Kluger, A.N. and 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), pp. 254–284. Available at: https://doi.org/10.1037/0033-2909.119.2.254 (Accessed: 21 September 2026).
• MindTools Content Team (2025) ‘The COIN Conversation Model’, Mindtools.com. Available at: https://www.mindtools.com/a94k5vp/the-coin-conversation-model/ (Accessed: 14 September 2026).
• Tomlinson, J. (2014) ‘BIF: A Simple Feedback Model’. Available at: https://johnrtomlinson.com/2014/04/22/bif-a-simple-feedback-model/ (Accessed: 14 September 2026).