Think about the last time you looked at your margin benchmark honestly.
Not the headline number. The real one. The figure that tells you whether this business can absorb a bad quarter, fund a senior hire, or hold its pricing in a competitive pitch.
It probably didn't shock you. You've seen something close to it before. The number moves a little between years but it doesn't break in either direction. And if you've suspected roughly what it would be for a while, and it still hasn't shifted. The question isn't what the number is. The question is what's holding it there.
The sector data doesn't make it easier.
Only 27% of UK agencies believe they are paid a fair price for their work. 58% report little to no progress reforming commercial agreements. The IPA first identified this structural pricing failure in 2020. Three years before ChatGPT launched.
That isn't a recent problem. It pre-dates the AI pressure entirely.
The pipeline picture compounds it. The Wow Company's BenchPress survey found that 46% of agency owners now name winning new business as their single biggest problem. Two years ago, that figure was 27%. Agency confidence has dropped to 66, an all-time low since the survey began tracking in 2012.
And then the AI pricing dimension lands on top of it. 27% of agencies have already been asked by clients to lower their prices because they use AI. 58% of brands expect to pay less for agency services where AI is deployed. Not in the future. That pressure is already in pitches, in contract reviews, in quarterly business reviews.
These numbers confirm what most founders already sense. They don't explain it. Something is connecting pricing difficulty, pipeline pressure, and AI fee erosion. The benchmarks show the outcome. They don't show what's producing it.
Here's what the benchmarks don't show.
They measure outcomes: margins, pipeline, pricing power. They don't capture the operational pattern producing those outcomes week after week.
When different people on the team use AI differently, the efficiency gains don't reach the P&L. They disappear into the workflow. One person uses a tool to halve a drafting task. Another sends client data into a platform nobody sanctioned. A third spends twenty minutes cleaning up AI output that wasn't ready for review. Nobody decided that. It just happened.
The time savings are real. The margin capture isn't.
And it compounds. Review habits differ by person. Standards for what "done" looks like vary by project. Assumptions about which client data can go into a prompt aren't written down anywhere. The team is using AI. The team is not using AI consistently.
That gap is invisible until someone maps it. And until it's mapped, you're managing the symptoms, not the cause.
It shows up in three recognisable ways.
The first is absorbed efficiency. The team completes tasks faster. The time saved gets filled with more work, more revisions, more scope that wasn't scoped. The P&L doesn't follow.
The second is the client question. Someone asks, in a pitch or a quarterly review, how the team uses AI. Different people in the room give different answers. Not wrong answers. Just different ones. That isn't a communications problem. It's a consistency problem.
The third is invisible justification. AI is producing real gains. But when pricing gets challenged, nobody can point to where those gains went. The efficiency is real. The evidence isn't.
Different surfaces. Same condition underneath.
The instinct is to add something.
A policy document. A tool audit. A training session. These are reasonable-sounding moves. They share one problem: they assume you already know what's happening across the team.
You can't govern what you haven't mapped. Policy built on assumption doesn't reduce inconsistency. It documents the assumption and calls it progress.
Before tools, training, or policy: find out what's actually happening.
Which functions are using AI. How. With what client data. How outputs are being reviewed before they leave the agency. That picture doesn't take months to produce. Two weeks is enough.
And it tells you what actually needs fixing, rather than what sounds right to fix.
If the consistency gap is live in your agency, happy to compare notes.
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. Subscribe.