Difficult Conversations at Work: What to Say When There Is No Good Version

TL;DR

• The core decision: whether you're going into the conversation to deliver the message or to be comfortable afterwards, because you can't optimise for both.
• When doing nothing is right: when the thing that's bothering you happened once, had no consequence, and you'd struggle to say what should change.
• What has to be true: you can state the message in one sentence, and you know what you want to be different afterwards.
• How the options split: by how the first minute goes, since that's where most of these are won or lost.
• Decision rule: if you can't say the message plainly in the opening minute, you aren't ready, and more preparation on the middle won't help.
• Outcome to expect: a worse hour and a better month.

The Conversation You've Been Rehearsing in the Shower

Something needs saying. It's been true for a few weeks now, and it's gone from an observation to a pattern to something that's affecting other people. You know it needs a conversation, and you've had the conversation several times already, in the car and in the shower and at three in the morning, where it goes well and the other person responds reasonably and you're both fine afterwards.

You haven't had it in the room. Instead you've done the smaller things: mentioned it obliquely in a meeting, hoped they'd pick it up from a change in your tone, half-raised it and then softened it into a question about workload. None of that worked, because none of it was the message, and now the thing has been going on long enough that the first question they'll reasonably ask is why you didn't say anything sooner.

Most advice on this subject is about the words. Structures, phrasings, ways to make the message land gently. Some of that helps at the margin. It doesn't touch the actual difficulty, which is that you want two things that don't fit together: you want the person to change, and you want the conversation to end with them feeling fine about you. Every technique for softening the message is a small trade of the first for the second, and enough of those and the message doesn't arrive at all.

Recommended Tools

Best tools for Leadership & Culture

C Culture Amp Culture measurement L Lattice Leadership development L Leapsome People enablement platform

View all recommended tools

The conversations that go badly usually went badly in the opening minute, before any technique had a chance to apply. That's also the part you can prepare properly, which makes it the most useful place to spend your attention.

When You Genuinely Do Not Need to Act Yet

Your current setup is genuinely fine. The thing happened once, nothing came of it, and if you try to state what should be different you find you can't. Not every irritation is a message. Raising a one-off with no consequence attached tends to land as disproportionate, and it spends credibility you'll want for something that matters.

Friction is starting to show. It's happened more than once and you've started adjusting around it: routing work differently, reviewing more closely, warning somebody else. That adjustment is the real signal, because it means you've already decided there's a problem and chosen to manage it silently. The cheapest check is to ask what you're doing differently because of this, and whether the person knows.

It has become a real cost. Other people have noticed, or the work is suffering, or you're avoiding the person. At this point the conversation is overdue and every week of delay makes it harder, because the reasonable question of why now has a worse answer each time. Waiting also removes the option of treating it as a small correction.

The edge case that forces it. Something has happened that can't be left: a serious incident, a complaint, or behaviour that affects somebody else's ability to work. In those cases there's usually a process involved, and the rules governing warnings, records and dismissal differ considerably by jurisdiction. Find out what applies where you are before the conversation rather than after it, and take local advice rather than following a template from elsewhere.

Five Questions This Reader Asks at 11pm

How do I start? With the message, in the first two sentences, in plain language. The instinct is to build context first so it lands more gently, and it does the opposite: the other person knows something is coming, and every sentence of preamble increases the dread while telling them nothing. Say what this is about, then explain.

Won't being direct damage the relationship? Less than the alternatives. People generally recover from being told something clearly. What damages relationships is finding out later that you'd been thinking this for months, or being managed around without being told, or receiving a message so hedged that they left the room without realising they'd been given one.

What if I'm wrong? Say so in advance, and mean it. Describing what you've observed and asking whether you've read it correctly is honest, and it's not the same as softening. What's not honest is using uncertainty as a way to avoid stating the message, so that if the conversation gets difficult you can retreat into having only asked a question.

What if they cry, or get angry? Both are common and neither means you've done it wrong. The mistake in each case is to start retracting: in the face of distress, managers reflexively soften the message until it dissolves, and in the face of anger, they either match it or capitulate. What both reactions actually need is a pause, and then the same message again, unchanged and unhurried.

Should I put it in writing? Depends what it is. A note afterwards recording what was discussed is usually sensible and protects both of you. Leading with a written message is different: it's appropriate for something formal, and it's a way of avoiding the conversation when used for anything else. If it's serious enough to need a record, find out what your jurisdiction requires before you write anything, and take advice.

Three Honest Categories the Approaches Split Into

Soften it to protect the relationship. You wrap the message in context, positives and qualifications so it's easier to hear. It's right in a narrow case: when the relationship is genuinely fragile and the issue is minor, softening buys you a hearing you might not otherwise get. It fails in the ordinary case because softening trades message clarity for your own comfort, and the trade is usually bad. The common result is a person who leaves the conversation believing they've had a chat about workload, carries on exactly as before, and is then genuinely blindsided when it turns out to have been a warning.

State it plainly and early. You open with the message, then provide the evidence and the conversation. It's right almost always, and it's the approach that respects the other person most, though it rarely feels that way in advance. It gives them the information they need to respond properly and it removes the excruciating opening where they're trying to work out what's happening. It fails when plainly is confused with bluntly: stating a message clearly doesn't require stating it coldly, and a flat delivery of something difficult reads as contempt whether or not it was meant that way.

Hand it to a process. You use a formal mechanism, a written note, a structured meeting with another party present. It's right when the matter is serious, when a record genuinely matters, or where local rules require a particular procedure, and it's right when there's a real power imbalance that makes an informal conversation unsafe for the other person. It fails when it's reached for too early, because escalating something that could have been a conversation removes any chance of an easy resolution and tells the person you were unwilling to speak to them directly.

Five Diagnostic Questions You Can Self-Assess Against

Can you say the message in one sentence? Write it down, out loud if you can. If it takes a paragraph, you haven't isolated it yet, and the conversation will wander until the other person constructs their own version of what you meant. The sentence doesn't need to be the first thing you say, but you must have it before you go in.

What do you want to be different afterwards? Be specific and observable. "Be more collaborative" isn't a request, it's a mood. "Tell the team before changing a shared deadline" is something a person can do on Thursday. If you can't complete this sentence, the conversation is about your frustration rather than about a change, and it will feel that way to them.

How long has this been going on? Count honestly, because they will. The length of time you've known is the most awkward fact in the room and it's better acknowledged than defended. If it's been months, say so, and say why you didn't raise it earlier, even if the honest answer is that you found it difficult.

Have you already decided the outcome? There's a real difference between a conversation where the decision is open and one where it's made, and pretending the second is the first is the most damaging thing you can do here. If you've decided, say what you've decided. People will accept a decision they dislike far better than they'll accept a consultation they later discover was theatre.

What's the version of this where you're the problem? Steelman it before you go in. Maybe the brief was unclear, the workload was unreasonable, or something you did made this the sensible response. You don't have to agree with that version, but you do need to have considered it, because if it's true and you haven't, the conversation will collapse in the first two minutes.

Five Ways Managers Open a Hard Conversation, Reviewed

The sandwich

Positive, then the message, then positive. It earns its place in exactly one situation: with somebody so fragile or so new that an unmodulated message would produce shutdown rather than change, where the opening positive is buying you a functioning listener. Even then it works only if the positives are true and specific.

Where it falls short is that it's the most widely recognised structure in this subject, and recognition destroys it. Anybody who has been managed before knows what the opening compliment means, so instead of softening the message it creates dread and devalues the praise. Worse, it gives a person who'd rather not hear the message a legitimate escape route: the conversation contained two positives and one criticism, and memory is generous.

The residual value of the idea is worth keeping in a different form. Say the true positive thing at a different time, in a different conversation, where it can be believed.

The impulse behind it is worth defending even though the structure isn't. Managers reach for this because they want the person to leave intact rather than diminished, and that's a decent instinct badly implemented. The thing that actually achieves it isn't a compliment bolted to either end of the message. It's making clear, once, that you're raising this because you expect them to be here and doing well, which is a statement about the future rather than a softener on the present.

The slow build with context first

You explain the background, describe the situation, establish the shared understanding, and arrive at the message once everything is in place. It earns its place when the message genuinely doesn't make sense without context, which is rarer than it seems, and when there's a decision to reason through together rather than a message to deliver.

It falls short because of what happens in the listener. They know from the invitation, from your tone, from the room, that something is coming. Every minute of preamble is a minute of anticipation without information, and by the time you arrive at the point they've usually constructed something worse than what you were going to say. Some of them will interrupt to ask you to get to it, and they're right to.

Context is valuable and belongs after the message rather than before it. Say what this is about, then take all the time you need to explain.

The direct opening

The first two sentences state the subject and the message. Something like: I want to talk about how the handover went last week, and I think it caused a problem we need to fix. It earns its place as the default. It removes the dread, it gives the other person the information they need to engage, and it signals that you'll be straightforward with them, which pays off in every subsequent conversation.

It falls short in one predictable way, which is that people confuse it with being harsh. Directness is about structure rather than tone: the message comes first, and it can still be delivered with obvious care. The other failure is directness without preparation, where a manager opens plainly and then can't substantiate it, which is worse than not opening at all.

This is the approach to default to, and the discipline it requires is entirely in the preparation. If you can't open plainly, it's because you haven't isolated the message yet.

What people usually find, having tried it once, is that the conversation is shorter and less awful than the rehearsed version. Most of the dread in these situations belongs to the anticipation rather than the event, on both sides, and a direct opening collapses the anticipation immediately. The person who has been waiting all morning to find out what this is about gets to stop waiting in the first fifteen seconds, and whatever they feel after that, it's a relief compared with not knowing.

Asking them to self-assess first

You open by asking how they think it went, or how they feel things are going, and let them raise it. It earns its place when there's a genuine chance they already know, which is often, and it's efficient when it works: somebody who names the problem themselves is considerably more likely to act on it.

It falls short in two ways. If they don't raise it, you've now got a worse opening than you started with, because you have to contradict the account they just gave. And if they know that you know, the question reads as a test with a right answer, which is a small humiliation dressed as a conversation. That's the version people remember years later.

Use it when you'd genuinely be interested in their assessment whatever it contains. Don't use it as a way to make them deliver your message for you.

There's a narrower version that avoids the trap. Ask the question, listen to the answer, and if they don't raise the thing, say so plainly rather than circling: that's useful, and there's something else I want to add. That keeps the benefit of hearing their account first without leaving you stranded when the account doesn't contain what you needed it to.

Putting it in writing beforehand

You send the substance in advance so they can absorb it privately and come prepared. It earns its place for anything formal, for people who process better in writing, and where the volume of detail would be hard to take in verbally. It also removes the ambush quality that makes some conversations unproductive.

It falls short when it replaces the conversation rather than preparing it. Written messages are read at the reader's worst moment and in their least generous interpretation, with no tone and no chance to ask a question. It's also the most common way managers avoid the harder thing, and that's visible: a difficult message delivered only in writing tells the recipient you weren't willing to say it to them.

Use it as a companion, not a substitute. And where the matter is serious, check what your jurisdiction requires of written records before you create one.

One detail changes how well this works, and it's the timing. Sending the substance the evening before gives somebody a night to sit with it alone, which is the worst possible use of the format. Send it a couple of hours ahead, during the working day, so they can read it, react privately, and bring their response while there's still somebody available to talk to. The point of advance notice is preparation rather than endurance.

The Decision Table

The row that costs most is the second, and it's the most common. Managing around somebody without telling them is a decision made in a series of small moments, none of which felt like a decision, and it compounds: by the time the conversation happens, the person has to absorb both the message and the discovery that you've been working around them for months.

Preparing the Part You Can Control

The middle of a difficult conversation can't be scripted, because it depends on what the other person says. The opening can be, and so can four other things. All of them take fifteen minutes, and skipping any of them is where improvisation causes damage.

The third row is where most preparation stops short. Managers usually arrive with the message and the evidence and no clear ask, so the conversation ends in agreement about a problem and silence about a solution. Both people leave feeling it went reasonably, and nothing changes, which guarantees a second and harder conversation later.

The fifth row is the one that saves conversations. If part of this is your doing, saying so early costs almost nothing and buys an enormous amount, because it converts the conversation from an accusation into a shared problem. If you wait for them to raise it, you're defending rather than leading.

When It Goes Sideways

Four reactions cover almost everything, and each one needs a different response. What they share is that the right move is slower than your instinct.

Distress. Someone becomes upset, and the reflex is to comfort, which in practice means retracting. Pause properly, offer a break, and be kind about the upset itself. Then repeat the message, unchanged. The kindness and the message are separate things, and giving up the second to provide the first is what produces the conversation that was had and changed nothing.

Anger. They push back hard, or make it about you. Don't match it and don't fold. Let it run for a moment, acknowledge it plainly, and return to the specific evidence: not a general claim about their attitude but the thing you observed on the day it happened. Specifics are harder to argue with than characterisations, which is the main reason to have prepared them.

Deflection. The conversation turns to somebody else's failings, to the process, to how impossible the constraints were. Some of it may be true and worth addressing separately. Say that explicitly, write it down in front of them, and then come back. The technique that works is acknowledging the point rather than dismissing it, followed by a return to the thing you came to discuss.

Total agreement. The easiest to mishandle, because it feels like success. Somebody who agrees immediately and fully may have understood, or may be ending the conversation as fast as possible. The check is specific: ask what they'll do differently and when, and listen for whether the answer is concrete. If it's a general commitment to do better, you've had a pleasant conversation rather than an effective one.

What to Put in Writing

Very little of this belongs in a file, and over-documenting a minor issue changes what it is. Three things are worth recording.

The second row protects both people, and sharing it matters as much as writing it. A record the other person hasn't seen is a file being built; a note sent to them the same day saying here's what we discussed and what we agreed is a normal professional courtesy. Where the matter is formal or may become so, requirements for what must be recorded and shared differ by jurisdiction, so establish that before you start rather than afterwards.

Questions to Ask Before You Commit

On the message. What is this about in one sentence? A bad answer needs a paragraph.

On the ask. What should be different, in a way somebody could act on this week? A bad answer describes an attitude.

On timing. How long have you known, and what will you say when they ask why now? A bad answer is that it has built up.

On your part. What did you contribute to this? A bad answer is nothing.

On the decision. Is the outcome open or settled? A bad answer is that you will see how the conversation goes, when you already know.

On the route. Does this need a formal process, and have you checked what applies here? A bad answer copies what was done last time.

What Getting This Wrong Costs

The first cost is the delay itself, and it's the one that quietly does the most damage. A problem raised early is a small correction; the same problem raised six months later is a pattern, a case, and a conversation with a defensive question built into it. Nothing about the underlying issue got worse in that time, but the conversation did, and the person's reasonable sense of unfairness is entirely a product of your postponement.

The second cost is that you become unreadable, and that spreads beyond the person involved. Teams work out quickly whether a manager says what they think. Where they don't, everybody starts interpreting: a slightly cool response becomes evidence of something, praise gets discounted as possibly strategic, and people spend energy trying to read a situation that could have been stated. The manager experiences this as a team that's oddly anxious, and doesn't connect it to the conversations they didn't have.

The third cost lands on the person who was never told. Somebody managed around for a year, given progressively less interesting work and never told why, has been denied the chance to fix something. When it finally surfaces, usually in a review or a decision about a promotion, their reaction isn't really about the feedback. It's about the year, and it's justified.

So before you go in, work out which of three things this is. Getting that wrong is the most common reason a well-prepared conversation still lands badly, because the person hears an accusation where they needed support, or a supportive chat where they needed a clear instruction. A behaviour problem means somebody is doing something they could stop, and needs a clear message and a specific ask. A capability problem means they can't currently do what's required, and needs a different conversation about support, time and whether the role fits. A clarity problem means they were never told what was expected, in which case the conversation isn't difficult at all, it's overdue information, and the honest version starts with an apology.

When You Are Ready to Go Further

None of this needs training. It needs the message written in one sentence, a specific ask, and an honest answer about how long you've known.

The step beyond your own practice is organisational, and it's where the real variation lives. In most companies, whether somebody hears about a problem early or a year late depends entirely on which manager they happen to have, and nobody can see that variation from inside a single team. Making it consistent is a question of what managers are expected to do and how that expectation is checked, rather than a question of anybody's courage.

HROpsLab publishes independent comparison work across HR tooling, applicant tracking and payroll. We sell nothing, we take no vendor money, and we publish no paid placements. If the next step is looking at how consistently these conversations happen across your organisation, our comparison work is one place to start.

Frequently Asked Questions

You can have the message land or you can be comfortable. Most of the damage in this subject comes from trying to have both.