The agency is busier than it has been in two years.
The team is producing faster. Status updates that used to take forty minutes are taking ten. Briefs are assembling in half the time. The account lead finishes Thursday afternoon with two clear hours back in the day.
At the end of Q1, the margin number is flat. Not bad. Not a crisis. Just flat. The same as last year, when nobody had these tools.
Most agencies assume AI is helping because the team is moving faster. That assumption skips a question: faster doing what, for whom, and does the saving land anywhere?
Here is one project, walked forward.
The account lead records a client call on Tuesday morning. The AI summary lands in her inbox. It is clean, confident, and structured. It is also, in two places, more certain than the call was.
The client said they "might want to explore" an additional deliverable. The summary reads: client has indicated interest in expanding scope to include a regional adaptation.
The account lead reads it quickly. It sounds right. It goes into the brief.
Strategy picks up the brief on Wednesday. It looks complete. Work begins.
By the following Monday, a full draft exists for a deliverable the client has not actually confirmed. The client sees it on Tuesday's call, hesitates, then asks for revisions on the core campaign instead. The extra work goes in the drawer.
The account director is pulled into a call to smooth the relationship. That call takes an hour. It is not logged anywhere.
The revision request that follows is absorbed. Scope was never formally agreed, so it is not formally broken. The account lead records it as a normal round.
Nobody has done anything wrong. The AI tool saved ten minutes on the brief. The project spent four hours recovering from the assumption the summary introduced.
The ten-minute saving did not disappear. It moved downstream, distributed across the project in pieces too small to track.
The saving had nowhere to land.
When the summary went straight into the brief, nobody decided what status to give the parts that were not yet confirmed. There was no field for assumed versus agreed. There was no step between "AI drafted this" and "Strategy inherited it."
Speed without a decision trail is not productivity. It is margin moving somewhere nobody is looking.
The same project, with one decision made differently, looks like this.
The account lead reviews the AI summary before it becomes the brief. It takes four minutes. One field is added to the handoff: open items, awaiting client confirmation. The regional adaptation goes in that field.
Strategy sees it. They note the dependency before starting. They do not build the extra deliverable.
The client call on Tuesday confirms the core brief. The regional adaptation is deferred to the next project cycle.
There is no recovery call. There is no absorbed revision. The account director's hour is available for something else. The saving from the AI summary is still a saving.
Same tool. Same account lead. Different moment. The thirty seconds it took to mark one item as unconfirmed.
The before and after are not separated by a better tool, a training programme, or a new policy. They are separated by one decision, at one handoff point, about where human judgement was needed before the work moved forward.
That decision does not happen automatically. Someone has to own it. It has to be visible in the workflow, not assumed.
Most agency founders I speak to know this pattern exists somewhere in their projects. What they usually cannot tell me is how often it happens, where it is worst, or how much it costs across a quarter.
A useful question to ask this week: when your account team uses AI on meeting notes or briefs, does the next person in the workflow know which parts are confirmed and which are still open?
If the honest answer is "probably not always," it may be worth looking at where the hours are actually going.
Craft with Command goes out weekly. If you are building a UK agency and want to think more clearly about how AI fits the work, you are in the right place.