How should my agency use AI in delivery?

The agencies building genuine AI capability into their delivery model are gaining a real cost advantage: better margin on the same revenue, because fewer hours are needed for equivalent or superior output. The absence of credible AI integration is increasingly visible to clients in pitches. So the question isn't whether to do this. It's how far to take it.

Clients buy outcomes, not effort

First, a mindset shift. There are still agency leaders who believe clients want work done by hand to keep standards high. Here's the reframe. Clients don't care how you get to the outcome. None of them. They care about whether it works, they care about what it costs, and then they care about whether it makes them look good.

Methodical delivery might feel boring next to magical creativity, but method is what proves you can deliver outcomes again and again. Sell the method, then dazzle them with the creativity.

The three levels

Ad hoc tool use. Individual team members using AI informally. No shared standards, no measurement, no integration into process. The efficiency gains are inconsistent and uncaptured, and this level is now expected as a baseline in most categories. It differentiates nothing.

Embedded workflow integration. AI built into specific delivery workflows, with shared standards, prompt libraries and quality checks, and with the impact measured. This is where consistent efficiency gains start flowing through to margin, and where the capability becomes demonstrable in a pitch.

AI-native delivery. Delivery processes designed around AI capability from the start, producing outputs that weren't possible before, with IP developing around the methodology. This decouples output volume from headcount and opens up premium pricing. It's comparable to what the networks are investing in, at a fraction of the cost.

Two conditions for keeping the gain

One: your commercial model. On time and materials, every hour AI saves is handed straight to the client. Pair your AI programme with a shift towards outcome or deliverable-based pricing, or you'll do the work and give away the benefit.

Two: effectiveness first. I talk a lot about efficiency, but never at the expense of effectiveness. You have to be effective first, then work out how to be efficient at being effective. AI applied to work that doesn't move the client's business just produces the wrong answer faster.