The five conversations agencies most avoid are: telling a client a project is late, raising scope that has crept, chasing an unpaid invoice, addressing results that are not where they should be, and ending a relationship. All five follow the same structure - state the situation plainly, take the part of it that is yours, present a plan, and name what you need from them. The single biggest determinant of how these go is timing: a problem raised early is a conversation, and the same problem raised late is a dispute. Delay is what converts one into the other, and delay is almost always driven by the hope that the problem will resolve itself.
Every agency has a conversation it is not having.
The project that is going to be two weeks late and nobody has said so. The client whose requests have quietly doubled the scope. The invoice that is 50 days overdue and has been mentioned twice, politely, in the last paragraph of unrelated emails.
These conversations are avoided for an entirely rational reason: they are unpleasant, and there is always a plausible story in which they become unnecessary. The team might catch up. The client might stop asking. The invoice might just arrive.
They rarely do, and the cost of waiting is not linear. A delay raised in week three is a scheduling conversation. The same delay discovered by the client in week seven is a trust problem, and trust problems are much more expensive.
This guide covers the general structure that works for all of these, then the five specific conversations with what to actually say.
Why delay makes everything worse
Worth being precise about the mechanism, because it is what motivates doing any of this.
When you raise a problem early, you are giving the client information plus options. They can move a date, reduce scope, add budget, or accept the impact. They are a participant.
When they discover it late, you have given them a fait accompli. The options have expired, and the only thing left is a reaction. You have also, implicitly, told them you knew and did not say - and that inference does more damage than the original problem, because it changes what they believe about every future update.
That is the whole argument. Early, you are a supplier managing a project. Late, you are a supplier who withheld something.
The structure that works for all five
Four parts, in this order. It is not a script - the order is the useful part.
1. State the situation, factually, first. No preamble, no build-up. "The build is going to be two weeks late" as the first sentence, not the fourth paragraph. Burying it reads as evasion even when it is only discomfort.
2. Take the part that is yours. If you underestimated, say so. Specifically, once, without over-apologising. "We underestimated the integration work" is credible. Three paragraphs of contrition is not, and it makes the client manage your feelings rather than the problem.
3. Present a plan. This is what separates a professional problem from an incompetent one. Never bring a problem without at least one proposed path. Two options is better - it gives them agency and signals you have thought about it.
4. Name what you need. A decision, a date, an approval, a payment. Be explicit and give it a deadline, or the conversation ends in vague mutual goodwill and nothing changes.
Two things to avoid throughout:
Do not lead with the explanation. Why it happened matters less than what happens now, and explanation-first reads as defence. It also invites the client to evaluate whether your reason is good enough, which is not a productive discussion.
Do not apologise more than once. The first apology is professional. The third makes the client uncomfortable and shifts the emotional labour to them.
The five conversations
1. The project is going to be late
Raise it the day you know, not the day it becomes certain. This is the hardest discipline on the list because there is usually a window where it might still be fine.
Rule of thumb: if you would be surprised to hit the date, say something.
The build is going to be about two weeks later than planned - we're now looking at 28 November rather than the 14th.
Two things caused it. The CRM integration was more involved than we scoped, which is on us - we should have checked the API before committing to the date. And the content for the three product pages arrived on the 9th rather than the 1st.
Two options. We can hold the full scope and move to the 28th. Or we can launch on the 14th with the product pages as placeholders and add them the following week - the rest is ready. My recommendation is the second, because it gets the site live before your campaign starts.
Can you let me know which by Thursday so we can plan the sprint?
Note the structure: news, ownership including the part that is theirs stated neutrally, two options with a recommendation, and a decision with a date. The client-caused delay is mentioned factually and not used as a defence - which is what makes it credible rather than defensive.
2. The scope has crept
The most avoided conversation, because each individual request felt too small to raise. Which is exactly why it is now a big conversation.
The critical move is evidence, not feeling. "It feels like we're doing a lot of extra work" is a position the client can disagree with. A list is not.
I want to flag something before it becomes a bigger problem.
Since we agreed the scope, we've taken on eleven requests outside it - the blog templates, the second language, the analytics dashboard and eight smaller ones. Individually they were all small and we absorbed them. Together they're about 60 hours, which is a third of the original project.
That's on me for not raising it sooner - I should have flagged it at the third or fourth one.
Here's what I'd suggest. We treat the eleven as a change order at £X, which covers the work already done and the two still outstanding. From here, anything outside the scope gets a quick estimate before we start, so you can decide each time rather than finding out at the end.
Does that seem fair?
Two things make this work. Taking responsibility for not raising it earlier is honest and it removes the accusatory edge. And proposing a process change means the conversation happens once rather than every month - our guides to change orders and preventing scope creep cover the mechanism.
If you are going to write off some of it as goodwill, show the full value and then discount it explicitly. Absorbing it silently teaches the client that out-of-scope work is free, which is how you got here.
3. The invoice is overdue
Agencies are strangely apologetic about being paid, and it is counterproductive. You delivered work under agreed terms. Asking for payment is administration, not an imposition.
The mistake is escalating too slowly and too vaguely. A structured sequence:
Day 1 after due date - automated, neutral. A reminder from the system, not a person. No emotional content.
Day 7 - short and personal, to your day-to-day contact. "Invoice 1042 was due on the 3rd - could you check where it is in your process? Happy to resend if it's gone astray." Assume administration, not refusal. Most late payments genuinely are administrative.
Day 21 - to the contact, naming the consequence. "This is now three weeks overdue. I need to flag that we'll have to pause work on the current phase if it isn't resolved this week - not something I want to do mid-project."
Day 30 - to whoever holds the budget, and pause the work. The pause matters. An agency that threatens a pause and does not implement it has taught the client that the terms are decorative.
Two things to get right in advance, because they are much harder to introduce mid-dispute: payment terms in the contract, including what happens when they are missed, and a named finance contact at the client, obtained during onboarding. Chasing through a marketing manager who has no visibility of the payment run is slow and puts them in an awkward position.
4. The results are not what you both hoped
Specific to outcome-oriented work - campaigns, conversion, SEO - and the hardest to handle well, because the temptation to hide behind metrics that did look good is enormous.
Do not do that. A client who later realises you highlighted impressions while conversions fell will not trust a report from you again.
I want to be straight about where we are. Three months in, conversions are up 8% against the 25% we targeted. That's real progress and it's well short of what we set out to do.
Here's what we've learned. The traffic improvements worked - sessions are up 40%. The conversion step is where it's stalling, and the data points at the form, which is longer than the benchmarks for this sector.
I'd suggest we shift the next two months from acquisition to the form and the checkout flow. That's a change of emphasis rather than of budget.
I'd also understand if you want to review the arrangement at the three-month mark rather than six. Happy to have that conversation.
The last paragraph is the one people leave out, and it is the most powerful. Offering the client an exit when results are behind is counterintuitive and it does two things: it demonstrates you are confident enough to survive the question, and it very frequently produces the opposite response. Clients rarely leave a supplier who volunteers accountability.
5. Ending the relationship
Sometimes you are the one who should end it - the account is unprofitable, the behaviour is unacceptable, or the fit is wrong.
Be direct, brief and non-negotiable, and give real notice.
I've been thinking about this carefully and I don't think we're the right agency for you any more. The work has moved a long way from what we're strongest at, and you'd be better served by a team that specialises in it.
I want to end this properly. We'll complete the current phase, deliver everything by the 30th, and I'll put together a full handover pack. I'm also happy to recommend two agencies I think would be a good fit.
I've valued working with you and I want to be useful on the way out.
Do not itemise their failings. However tempting, however justified. The list will be forwarded, and it will be what people remember - not the two years of good work.
Do offer the handover and the referrals. Client offboarding done well after a difficult ending is disproportionately effective. The client already knows the relationship was not working; what they will describe to others is how you behaved when it ended.
Preparing for the conversation
Fifteen minutes of preparation changes the outcome more than any phrasing.
Write the first sentence down, verbatim. The opening is where these conversations are won or lost, and it is the part most likely to come out hedged under pressure. Writing it means you say "the build is going to be two weeks late" rather than "so, I wanted to touch base about timings."
Gather the evidence before, not during. Especially for scope and payment conversations. Hunting for the list mid-call weakens the position and makes it feel improvised.
Decide your concession in advance. What you are willing to give, and what you are not. Deciding this live, under emotional pressure, is how agencies end up writing off far more than they intended. Write the number down before the call.
Pick the channel deliberately. Anything with emotional content goes on a call, not in writing. Email is efficient for facts and terrible for tone - it strips out everything that signals goodwill and invites the reader to supply the worst plausible reading. Written summary after the call, always.
Choose the time. Not Friday afternoon, when it will sit unresolved over a weekend and grow. Not first thing Monday. Mid-week, mid-morning, with room in your diary afterward in case it runs long.
Handling the reaction
Some of these conversations will produce a strong response. Three things help.
Let the silence happen. After you state the situation, stop. The pause will feel much longer to you than to them, and filling it with additional explanation is the most common way a controlled conversation becomes a defensive one.
Do not match their register. If a client is angry, the instinct is to become either defensive or excessively apologetic. Neither helps. Staying level, factual and unhurried is what de-escalates - and it is much easier when you have a plan to offer, which is why the plan is step three of the structure.
Separate the reaction from the decision. Someone can be genuinely annoyed and still agree to your proposal. Do not treat frustration as rejection and start conceding to make it stop.
If it gets heated, break it. "I can hear this has landed badly, and I don't think we'll resolve it in the next ten minutes. Can we pick it up on Thursday when we've both had time?" This is not avoidance - it is recognising that decisions made in a heated conversation are usually bad ones, and it almost always improves the outcome.
When the client is wrong
Sometimes the problem is genuinely theirs - approvals that never arrive, contradictory feedback from unnamed stakeholders, a scope they keep expanding while insisting the date holds.
Three rules.
Raise it as a shared problem, not a fault. "The date is at risk and the main driver is approval turnaround - we've waited an average of nine days against the three we agreed. What can we do to speed that up?" This is factual and non-accusatory, and it produces a solution far more often than an assignment of blame does.
Show the arithmetic, not the grievance. "Each week of delayed feedback moves the launch by a week" is a statement about a schedule. "We're constantly waiting on you" is a complaint. The first gets a response.
Fix the mechanism, not the instance. If approvals are slow, the useful outcome is a named approver, a delegate for their absence, and an agreed turnaround written into the communication plan - not an apology for this one delay.
Getting the internal conversation right first
Before the client conversation, there is usually a team conversation, and getting that one wrong makes the other one much harder.
Establish the facts before assigning any of them. Teams under pressure narrate rather than report - "the client keeps changing their mind" is a narrative; "we received revised requirements on the 4th, 11th and 19th" is a fact. You need the facts, because that is what you are taking into the client conversation and what will not survive contact if it is wrong.
Decide who speaks, and speak with one voice. Nothing undermines a difficult conversation faster than a client hearing a different account from a designer than from the account lead. Agree the message, agree who delivers it, and make sure everyone with client contact knows what has been said.
Do not let the team discover the problem from the client. If you are about to tell a client the project is late, the team should already know that is happening and why. Being blindsided in a shared call is corrosive to trust internally in exactly the way withholding is externally.
Separate accountability from the client conversation. Whatever went wrong may need addressing internally. It should not be addressed in the same twenty minutes as deciding what to tell the client - the first is about learning and the second is about the plan, and mixing them produces a defensive team and a worse plan.
Four habits that reduce how often you need these
Say the small thing early. Most difficult conversations are the accumulation of easy ones that did not happen. The third out-of-scope request is an easy conversation. The eleventh is not.
Write down what was decided. Three bullets after every call, sent within the hour. Most "I thought we agreed" disputes are a memory problem rather than a bad-faith one, and a written record removes the category.
Agree the escalation route before you need it. A communication plan that names who raises concerns to whom makes escalation feel routine, which means problems surface earlier and smaller.
Make the data visible continuously. Most difficult conversations are about a gap between what the client believed and what was true - on progress, on hours, on retainer burn. A client who can see the state of things does not need to be told bad news, because they have been watching it develop and frequently raises it first.
The conversation you should have had a month ago
A closing note on the most common situation: you are reading this because a conversation is already overdue.
The instinct in that position is to wait for a natural opening, or for the situation to improve enough that the conversation gets smaller. Neither happens. What actually happens is that the delay itself becomes part of what has to be explained, which makes the conversation harder every week you leave it.
The way through is to name the delay as part of the opening rather than hoping it goes unnoticed:
I should have raised this three weeks ago and I didn't, which I'm sorry about. Here's where we are.
That single sentence does a surprising amount of work. It removes the thing you are most worried about them noticing, it demonstrates that you are being straight now, and it moves the conversation immediately to the substance. Clients are far more forgiving of a late disclosure that is owned than a late disclosure that is dressed up as timely.
Then run the same four-part structure - situation, ownership, plan, what you need - and do not spend any further time on why it was not raised sooner. One sentence is accountability; three paragraphs is asking the client to reassure you.
The one thing to take away
None of these conversations get easier with practice, exactly. What changes is that you have them sooner.
An agency that raises problems at week three has scheduling conversations. An agency that raises them at week seven has disputes. Same problems, same people, same underlying facts - and the entire difference is four weeks of hoping it would sort itself out.