The client asked for one more thing on a Tuesday call. It sounded like a tweak — could the export also handle the old format their warehouse system produces. You said you would look at it. By Friday you have eleven hours in it, the old format has three variants, and the fixed fee no longer covers what is left.
Now you have to tell them. That is the message freelancers rewrite four times and send at ten at night.
What the defensive version actually says
The usual draft opens with an apology, then spends three paragraphs on history — the Tuesday call, what was said, what you reasonably assumed — because you are trying to prove you were not careless. The number appears near the bottom, hedged: "this would probably add somewhere around another 3-4k, roughly."
Read it from the client's chair. The first sentence admits fault. The middle argues a case nobody made. The last one asks for money. So they answer "I thought that was included?" and now you are negotiating about blame instead of deciding about scope — a conversation that costs more than the eleven hours did, and one you started.
The defensiveness is a scheduling artifact, not a character flaw. You wrote it at the end of a day spent doing the work you are about to charge for.
Lead with the change and its cost
Put the change and the number in the first two sentences. Before the history, before the apology, before the reasoning.
"The legacy export format has three variants rather than one. Handling all three is about 25 hours beyond the agreed scope — $3,750 at our rate. Three options below."
That is not bluntness; it is the shape of a message that respects the reader's time. Two lines in, the client knows what this is about and what resolving it costs. Name the figure, too — "roughly" and "somewhere around" feel considerate and read as evasive.
A bill for the past invites an argument about the past
Two things are in your head and they must not share a paragraph. The first is the record: work done, hours spent, what was agreed and when. Factual, one short block. The second is the decision: what happens to the work that is left. That is the only part they act on.
Merging them produces the blame conversation, because a paragraph that mixes "here is what I already did" with "here is what I need from you" reads as a bill for the past — and bills for the past invite arguments about the past.
If some of the overrun is genuinely yours, say so in one sentence inside the history block and write off those hours explicitly. A specific concession ends an argument; a vague apology opens one.
Give them a bounded choice, not a problem
Do not send a problem. Send a short menu.
- Proceed at the revised cost. All three variants, plus 25 hours, delivery moves to the 12th.
- Hold the cost. Ship the one variant that was scoped; the other two become a later phase.
- Pause. I stop at the current milestone while you decide.
Each option costed in money and time, with your recommendation named. Only include options you would actually accept — one you are hoping they refuse reads exactly like what it is.
"Let me know how you would like to proceed" hands the client an open problem they know less about than you do, and the predictable outcome is silence, then a meeting. Asking for a decision has a general shape; this is that shape applied to money. Give a real deadline too, not a nervous one — inflated urgency spends credibility you will want later.
Say which numbers are logged and which are estimated
Almost every scope dispute traces back to one sentence: "you said it would be X." Usually you did say X, X was an estimate, and nothing marked it as one. In prose, a measurement and a projection look identical. So mark them:
- "11 hours logged as of Friday" — a fact, checkable against the timesheet.
- "About 25 hours for the remaining two variants — my estimate, and it assumes the three sample files you sent are representative" — a projection, with the thing it rests on.
The assumption matters more than the number. A client who knows those samples are unrepresentative can tell you now, before the hours are spent. An unlabelled estimate can only be judged afterward, when disagreeing has become expensive.
Where the assistant earns its place
Generating the message was never the problem. The tired version of you editing it into an apology at ten at night was.
RelayLink carries this as a briefing, and two fields do real work here. Options considered is part of the package, not something you remember at ten at night. And every assumption arrives labelled stated by sender or inferred by sender's AI, which makes an estimate visibly an estimate on arrival — so "you said it would be X" has something to point at. The draft also sits server-side until you confirm it, so what goes out can be the version you reread the next morning. More on the wider pattern in freelance client communication with AI.
What structure cannot do
None of this makes the conversation pleasant. A well-shaped scope message can still end with someone annoyed, negotiating hard, or cutting the phase — and sometimes cutting the phase is the right call.
What structure changes is what the conversation is about: facts, options, and cost, rather than whose fault Tuesday was. Smaller than it sounds, and the one actually on offer.
If the next one is already sitting in your drafts, connect your assistant and let it assemble the options while you decide which you can live with.