Most agencies are quite good at measuring output.
Turnaround times. Campaign performance. Content volume. Billable hours against the plan. AI has sped some of that up, and in plenty of cases the numbers look better than they did eighteen months ago. Drafts move faster. Decks come together quicker. Research takes a fraction of the time it used to.
But output speed isn't where the real question sits. The real question is about input. Not what came out, but how it got there. Which tools were used, on what, with what going into them, and who checked it before it went anywhere near a client.
That's a much harder question for most agencies to answer with any confidence.
This is where the five-level AI maturity ladder is useful. Not as a scorecard to chase, and not as a benchmark to compare yourself against other agencies. As a mirror.
Level one is individual use. People using AI for drafts, research, brainstorming, admin, off their own initiative. Useful, but personal and largely invisible to leadership.
Level two is task assistance. AI doing specific jobs inside existing workflows. Faster outputs on familiar work, but the underlying process hasn't really changed.
Level three is workflow copilot. AI embedded in repeatable workflows with defined inputs, outputs, handoffs, and human checkpoints. This is where consistency starts to show up, not just speed.
Level four is governed workflow intelligence. AI surfacing patterns and risks inside agreed ways of working. Data boundaries are clear. Review standards are clear. AI recommends, people decide.
Level five is ambient business intelligence, where the system is watching for friction, quality risk, and margin leakage without being asked. Most agencies are nowhere near this, and don't need to be yet.
Here's the part worth sitting with. Most agencies are somewhere between level one and level two. But plenty of founders would describe their agency as level three, because they've got tools rolled out, some training delivered, and a few people genuinely excited about what AI can do for them.
Tools and enthusiasm aren't the same as a workflow. Having AI available to the team is different from knowing, with any precision, where it's being used, on what kind of work, and what happens to that output before a client sees it.
The gap between where an agency thinks it sits and where it actually sits isn't a failure. It's information. It's the most useful single thing the ladder can tell you, because it points directly at where the structural work needs to happen next.
So bring it back to input. Pick any piece of client work that went out this month with AI involved somewhere in the process. Could you say, with a straight face, which tool was used, what was fed into it, and who reviewed the output before it left the building? Not in general terms. For that specific piece of work.
This is where it touches the actual work, not the policy document. A strategist using AI for early research has a different risk profile to a copywriter drafting client-facing language that needs sign-off, which has a different profile again to someone building a deck that goes straight to a regulated client. Three different functions, three different review points, and most agencies have never mapped any of them.
That's not a governance problem in the heavy sense. It's not about writing a policy nobody reads. It's closer to basic operating discipline: knowing where AI touches the work, what goes in, and who's accountable for what comes out. Governance as a habit, not a headline.
The honest answer for most agencies, if a regulated client asked tomorrow, is that different people in the building would give different answers. Not because anyone's doing anything wrong. Because nobody's actually mapped it yet.
That's the gap the ladder is pointing at. Not a future state to aspire to, just an honest read of where things stand now, function by function, piece of work by piece of work.
If that's ringing any bells for your agency, the AI Workflow Clarity Audit is built around exactly this question, working through where AI is actually being used across the team, what's going in, and where the review points are. Two weeks, fairly hands-on, and it leaves you with a clear picture of what needs tightening first.
If this is live for you at the moment, happy to have a twenty-minute conversation and compare notes. No pitch, just useful to talk it through with someone outside the building.
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.