Define Done Before You Start, or Argue About It Later
"Done" is the cheapest word in a project to define and the most expensive one to leave vague. Written down before the work begins, it costs ten minutes and one awkward question. Left undefined, it gets settled in week eight by whoever feels strongest, and the settlement is usually paid in unpaid hours.
The reason it stays vague is not laziness. It is that "done" feels obvious while you are planning and becomes contested only at the moment you need it most — when you are tired, the invoice is waiting, and the client says "almost."
Deliverables are not done
A deliverable list says what you will produce. A definition of done says what has to be true before you stop.
Those are different sentences, and the distance between them is where most revision fights live. "A market report" is a deliverable. "Three markets, one section each, every figure traceable to the appendix, ten pages of body text, revision round limited to factual corrections" is a definition of done.
The first can be produced and still be rejected. The second can be checked by someone who was not in the room.
Three definitions, rarely written down together
On any engagement there are at least three working definitions of done running in parallel.
Yours is when you have stopped improving the thing. You know the shape of finished work, and you can feel when another pass would be polishing rather than fixing.
The client's is when they can use it for the thing they had in mind — which is often something they never described, because to them it was obvious.
The invoice's is when the agreed object exists, in the agreed form, and the acceptance condition has been met.
When the three line up, delivery is quiet and nobody remembers it. When they don't, you get the sequence everyone recognises: you deliver, the client says "almost," you do another round, and the round produces small requests that were never in scope and are individually too small to refuse.
The fix is not to guess harder. It is to make your definition explicit enough that the other two have something concrete to disagree with — early, in writing, while disagreement is still cheap.
What a checkable definition looks like
Not a paragraph. A list of statements that a third party could mark true or false without asking you what you meant.
The weak version, which sounds professional and decides nothing:
"The report will be comprehensive, well-researched, and ready for the board."
"Comprehensive" is arguable. "Well-researched" is arguable. "Ready" is the most arguable of the three, because readiness is a feeling held by the person about to present it. Every word here is a judgement, and judgements cannot be inherited by a second reviewer.
The working version:
- Covers the three markets named in the brief, one section each.
- Each section states the recommendation and the two strongest reasons against it.
- Every figure traces to a source in the appendix; no unsourced numbers.
- Ten pages of body text, excluding appendix and cover.
- Delivered as a PDF and an editable document.
- One revision round, limited to corrections of fact and clarity — not new analysis.
Now disagreement has a shape. A fourth market is visibly outside the list, so it becomes a scope conversation instead of a quality complaint. A tone change sits ambiguously inside "clarity," which is exactly why the revision clause is written as a boundary rather than a number of rounds.
Notice what the list does not contain: aspiration. Aspirations cannot be checked, and anything that cannot be checked gets checked later against someone's memory of what they expected.
The four questions that produce the list
You do not have to invent the criteria. Four questions, asked before work starts, generate most of them.
1. What will you do with this when you have it? The answer fixes the real acceptance test. A deliverable destined for a board pack has a different length and format constraint than one handed to an engineering team.
2. What would make you send it back? Ask it plainly and write the answer verbatim. Clients answer this far more honestly than they volunteer it, and their answer is the closest thing you will get to their internal definition.
3. What is explicitly not part of this? Exclusions are half the definition. A list of inclusions alone can always be read as incomplete.
4. Who signs it off, and what do they need in front of them? Not who receives it — who says yes, and what evidence that person will want to see. A deliverable approved by a committee needs a different shape from one approved by a single operator.
Write the answers into one block in the proposal. Keep it to a page. This block is the thing you will point at later, so make it the thing you would want pointed at.
The revision clause is the boundary
Most delivery disputes are not about quality. They are about how many times quality gets to be redefined.
"One round of revisions" is a boundary. "Revisions as needed" is an invitation. "Unlimited revisions until you're happy" is a business model you did not intend to offer.
Two habits make the boundary hold. First, name what a revision round is allowed to contain — corrections of fact, corrections of clarity — and name what it is not — new analysis, new markets, new audience. Second, when the round is used, say so at the time: "That's the included round; anything further is a new piece of work and I'll quote it."
Saying it once, at the moment the round is consumed, is far cheaper than arguing about it at invoice time, when every sentence sounds like a negotiation.
Three failure modes worth naming
Done inflation. Late in a project the client starts adding small completion criteria — "it would be nice if it also had…" — each defensible alone. They arrive after the definition exists, which makes them feel like elaboration rather than scope. Treat them as scope. The sentence that works: "That's not in the list we agreed; I can add it for [amount] or swap it for [item]."
The signer changes. The person who approved the definition leaves, and their replacement has a different sense of finished. The definition still holds — but only if you re-confirm it with the new signer in writing, once, before the next delivery. Silence here is read as consent to a fresh definition.
The client who cannot be satisfied. No definition of done fixes a client who does not intend to pay or who needs the project to remain open for internal reasons. What the definition does is make that visible early, in a form that lets you stop. This is the outcome nobody enjoys and everybody needs.
What the definition does not do
It does not remove judgement from your own work. You still have to decide when the thing is good — the definition only makes that decision checkable by someone else.
It does not prevent a genuinely wrong scope. If the work truly needs to be twice the size, a precise definition exposes that in week two instead of week nine. Earlier exposure is the correct outcome even when it is the uncomfortable one.
And it does not replace the conversation. A definition nobody discussed is a document you will lose an argument with. The value is in the ten minutes of disagreement it forces while disagreement is still free.
Where a model helps
Converting a rough deliverable description into checkable statements is editorial work, and it is the kind a drafting tool does well — restructuring, not deciding.
A usable instruction: "Here is what I've agreed to deliver. Rewrite it as statements a third party could mark true or false. Flag anything that is still a judgement rather than a test."
The flagging is the valuable half. Anything the model cannot convert into a checkable statement is a place where you and the client are still relying on shared intuition, and shared intuition is what fails at handover.
The failure mode is worth knowing. Ask a model to write a definition from a thin brief and it will produce a confident, plausible list — including criteria nobody agreed to. Those criteria read like a fair standard, which makes them hard to refuse later, and they can commit you to work you never priced. The rule is the same as everywhere else: the model may restructure what you decided. It may not decide. Every criterion should trace back to a sentence in an email you can point at.
The handover is the test
The definition earns its keep at one moment: when you send the work and say it is done.
That message should not be "here's the report, let me know what you think." It should name the test: "This is the final version. Against what we agreed, it covers the three markets, includes the counter-arguments, sources every figure, and runs to ten pages of body text. The revision round covers factual corrections and clarity. If I've missed something on that list, tell me by Friday; otherwise I'll invoice on the agreed terms."
Two things can happen. The client confirms, and the delivery closes in one exchange. Or they name the specific item that does not meet the list — which is a real defect and worth fixing.
What stops happening is the open-ended "almost." An open-ended response has nothing to push against. A response to a written list does.
A ten-minute exercise
Take the last thing you delivered. From memory, write the list of statements that would have told you it was done. Then compare it with what the client actually pushed back on.
If the pushback landed on something in your list, the definition was fine and the problem was the conversation. If it landed on something absent from the list, that absence is the thing to fix on the next engagement — and it is much cheaper to notice now than in week eight.
This note describes a method. No time savings, completion rates, or client engagements are claimed. No figures appear in this article; where amounts are referenced elsewhere on this site they are illustrative placeholders, not market rates.
Related notes
- The Proposal Doesn't Need Writing. It Needs Deciding. — the exclusions block is where a definition of done actually starts.
- The Meeting Notes Are Not the Work. The Decision Log Is. — how an accepted definition survives contact with the following month.
- The Client Brief Is a Draft, Not a Specification — the acceptance test usually appears in the brief, unstated.
- Scope Doesn't Creep. It Gets Invited. — what to do when the boundary gets tested after it is written.