Every client approval process is tested by one line, spoken weeks after the yes: that is not what we signed off. The approval survives that line only when its record states what the client was shown, what the yes covered, and who gave it. The usual record is the yes and nothing around it.
The yes is the least informative part of an approval
When the dispute arrives, the account manager goes looking for the approval and finds it: a reply that says approved, sometimes with an exclamation mark. It is the one line in the whole engagement nobody argues about, and it settles nothing, because the argument is about what the client said yes to and never about whether they said it.
| What the dispute turns on | What the agency finds in the thread | What would have settled it |
|---|---|---|
| What the client was shown | An attachment somewhere in the thread | The artefact as it was sent, named in the yes |
| What the yes covered | The word approved | A line saying whether the yes covered the layout, the copy, the imagery or the lot |
| Who gave it | A name in the from field | A named approver, agreed before the first round |
| What was agreed to change first | A call | The list of changes, written back and confirmed before the yes was recorded |
| What changed after the yes | Nothing, because nobody logged it | The change, with its own approval |
Each row is a thing the agency could have written down at the time for the cost of a sentence, and each is a thing that cannot be reconstructed afterwards, because the client’s memory of it is the other side’s evidence.
Approved with changes is two approvals
The most common shape of a client approval is a yes with a list attached. Approved, but make the hero image warmer, shorten the intro and swap the second testimonial. Everybody hears that as a yes, and it is a no with instructions attached, one that becomes a yes only when the client has seen the version with the changes made and said so again.
The second yes is the one that goes missing. The agency makes the changes, the changes are small, and the project moves on. Weeks later the client sees the launched site, notices that the hero is warmer in a way they did not picture, and says that is not what we signed off. They are right. They signed off a description of a change, and nobody showed them the change.
The rule that follows is that a yes with a list is a round, not a close. The round closes when the client has seen the artefact with the list applied and said yes to that one. Where the changes are small enough that a second review feels like ceremony, the second review is a screenshot of the changed region and a one-line reply, and it still happens. Whatever it costs is less than the launch-day conversation.
Silence keeps a schedule and settles no argument
A deemed-approval clause in an agency contract treats a client’s silence past a deadline as approval. The clause exists for a good reason: a client who goes quiet cannot be allowed to stop the project, and as a schedule device it works. What it does not do is stop the argument. Whether the clause holds in a given contract is a question for a lawyer; what happens on the day is that a client who never looked at the artefact does not experience a deemed approval as having approved anything, and contests the result, whatever the clause says, at the moment the relationship is worth the most.
So the clause is used to move on, and the record is built as though it did not exist. When the deadline passes in silence, the account manager writes to the client that the round has closed under the clause, names the artefact now treated as approved, and says what happens next. That message is the record. A client who reads it and stays silent has now been told twice; a client who reads it and objects has reopened the round before the build started, which is the cheap moment to do it.
Approval by silence is never used for the artefacts a dispute is most likely to concern: the final design, any copy carrying a legal or pricing claim, and the launch itself. Those need a person to say yes, on the thing, in writing.
A change after the yes needs a yes of its own
The approvals that fail last are the ones that were sound when they were given. The client approved the design; then a developer changed a spacing rule to fix a wrap, a copywriter tightened a headline, and a client-side editor put content into the CMS that broke a layout the client had signed off with placeholder text in it. None of those people thought of themselves as changing an approved thing, and the account manager did not know most of it had happened.
An approval is a statement about a state. Once the state changes, the approval describes something that no longer exists, and the change is either re-approved or logged as a departure the client has agreed to. A list does this work: each change after the approval is a line, with who asked for it, what it touched and whether the client saw it. The list is short on a well-run project, and it is the document the dispute reads.
Where the client is reviewing the built site rather than a mockup, the record can produce itself. In Notarus, a change the client asks for on the staging site is a ticket on the team’s board from the moment they file it, with their original wording kept on it, and the client portal lets them watch that ticket reach Done and read the replies the team chose to show them, while the team’s own notes and priorities stay out of view. Each line of the change log this section asks for is then a ticket the client wrote themselves, and both sides hold the same copy.
The same act closes a web build. UAT sign-off is the client’s acceptance that the built site does what the acceptance criteria said it should, given against the site the client will get rather than against a description of it, and it is the last approval in the engagement that the words that is not what we signed off can be aimed at.
The record shows what the client was sent, no more
A record shows what the client was sent, when, and what they replied, and no tool extends it further. It can show at most that the artefact was opened; it cannot show that the client read it. A client who replied approved to a message skimmed on a phone will dispute the result with as much conviction as one who studied it, and the record’s job at that point is to establish what they were shown, which it does, rather than that they understood it, which nothing does.
The practical consequence is that the approval has to be made cheap to do properly. A long approval document nobody opens is worse than a short act on the thing itself: the client opens the artefact, sees what was sent, sees the list of what changed since last time, and replies to a question that names those things. The shorter that act is, the more likely the client did it, and the more the record is worth on the day somebody says that is not what we signed off.
