Hard conversations

Almost every management failure I've watched, including my own, was a conversation somebody postponed.

Not a bad decision. Not a missed strategy. A specific conversation with a specific person that everyone could see was needed, that got deferred for a plausible reason, and then deferred again because the deferral itself had made it bigger.

The skill is having it early, plainly, and once.

Early means the second time

"Early" needs a definition, because on its own it's the kind of advice that sounds right and gives you nothing to act on.

The first occurrence is an incident. People have bad weeks, misread a room, get something wrong. Escalating a single event into a Conversation is its own failure: it's disproportionate, it reads as surveillance, and it burns credibility you'll want later. Mention it in passing if it's worth mentioning at all.

The second occurrence is a pattern, and that's when it gets raised properly.

I hold the same rule for organisational change: wait until the same failure has happened twice or three times before building process around it. This is that rule pointed at a person instead of at a system, and it's the same reasoning. One event carries almost no information about what's structural. Two carries a lot. Waiting for five is not diligence, it's avoidance with a methodology.

The cost of waiting is specific and it compounds. Every week you don't say it, the person continues, which means they're accumulating a track record you're going to hold against them without having told them. The team is watching, and drawing conclusions about what's tolerated here. And the conversation itself gets harder, because now you're not raising a pattern, you're explaining why you didn't raise it in March.

Preparing without scripting

Prepare, because unprepared hard conversations wander into the general and the general is unanswerable. Don't script, because a scripted conversation is a performance the other person can feel.

Three things worth having ready.

The specific examples. Two or three, with dates, described as behaviour rather than as character. "In the architecture review on the 4th you cut Sam off twice and the discussion ended there" is workable. "You can be dismissive" is not, because it's a claim about who they are and the only available response is to defend themselves.

The one sentence. What is the thing you're actually saying, in a single sentence, that you'd be prepared to have quoted back. If you can't write that sentence, you're not ready, and you'll discover it mid-conversation and start hedging.

What you want to be different, concretely. Not "improve communication." What specifically, observable by when.

Then let the rest be a conversation, because the useful half is what you learn. A meaningful share of these end with information that changes your view: they're carrying something you didn't know, or the situation is a symptom of something structural, or you've been wrong.

Opening plainly

The message goes in the first thirty seconds, before any context, and the softening moves people default to are the ones that do the damage.

The compliment sandwich is the most common. Praise, criticism, praise. Everyone recognises the pattern, the praise is heard as setup and stops being believed, and half the time the message in the middle doesn't register at all. I've watched people leave those conversations genuinely believing they'd gone well.

The other one is the interrogative opening: "how do you think the project went?" It's an invitation for the person to convict themselves, and it's transparent. If you've decided something needs saying, say it.

What works is direct and specific and unhurried. "I want to talk about something I've noticed twice now, and it's affecting the team." Then the examples. Then stop, and let them respond, and actually listen to the response rather than waiting for it to end.

Plainly doesn't mean coldly. Warmth is fine and useful. What isn't fine is warmth that dilutes the content until the person walks out unsure whether anything was wrong.

The reaction you didn't plan for

You'll get one of four, roughly, and only one of them is the one people rehearse.

Acceptance, which is straightforward and slightly suspicious if it's instant. Ask what they'll do differently and listen for whether they've actually understood it or are just agreeing to end the conversation.

Defensiveness. Normal. Let it happen, don't argue point by point, and come back to the pattern rather than to the individual example. Arguing about whether that particular meeting really went that way loses the thread every time.

Emotion. Tears, anger, shutting down. Slow down, stop talking, offer to continue later, and mean it. What you must not do is retract the substance because the reaction was uncomfortable. That's the moment where most hard conversations get undone, and the person leaves believing it wasn't serious.

And information you didn't have. This is the outcome to actually hope for. If the reason is something you didn't know, say so, revise, and be visible about having revised. It costs you nothing and it's the thing that makes the next one possible.

Following up in writing

Send a short note afterwards, the same day. Three or four lines: what we discussed, what we agreed, when we'll check in.

The value is that memory of these conversations diverges enormously within a fortnight, usually in the direction of softness. You remember delivering a clear message. They remember a conversation about how things are going. The note prevents the drift, and it means the check-in starts from a shared record rather than from two accounts.

Keep it factual and short and don't restate the criticism at length. A long written summary reads as a file being built, which is a different message from the one you were trying to send.

If the situation is genuinely at the stage where a record matters, that's a different kind of conversation and it should be named as such, plainly, rather than achieved by tone.

Once

The word doing the most work in the whole thing.

Having said it clearly, don't say it again next week in a slightly different way, and don't keep circling back to check that it landed. That converts a piece of feedback into a campaign, and it tells the person that you don't believe they heard you.

Say it once, agree what's next, and then give them the space to change. Check in at the point you agreed, not before.

If nothing has changed by then, that's a new and much shorter conversation, and it's a different one: it's about whether this is going to work. Which is uncomfortable, and considerably more honest than a year of gentle repetition.