Why AI Isn’t Fixing Agency Margins (And What Actually Will)

Many agency founders are asking the same question right now: if AI is meant to make agencies more efficient, why do margins still feel so tight?

Work is coming in, teams are busy, and yet profit is harder to protect than it should be. Founder time is still heavily tied up in delivery, senior staff are acting as the safety net, and every month feels slightly more fragile than expected once costs are accounted for.

The problem usually gets blamed on pricing, but in most agencies pricing isn’t the root cause. The real issue is structural. It’s about how work flows through the business, where thinking still sits, and how much invisible effort is absorbed without ever being priced or designed properly.

AI hasn’t fixed that because AI, on its own, doesn’t fix broken operating models.

Busy agencies are not the same as profitable agencies

One of the most common misconceptions in agency businesses is that a busy team means a healthy business. In reality, many agencies are busy precisely because their operating model relies on people filling gaps.

Founder judgement is embedded in daily delivery decisions. Senior team members step in to fix issues late in the process. Rework happens because expectations weren’t clear upfront. Capacity looks fine on paper, but collapses in practice once complexity, client nuance, and interruptions are factored in.

This is where margin quietly disappears. Not in one big mistake, but through dozens of small, repeated inefficiencies that compound over time.

Why AI usually doesn’t change anything

Most agencies will say they are “using AI”, but when you look closer, that usually means limited, individual use rather than operational change. Someone might use AI to draft content, summarise notes, or speed up research, but the underlying delivery model stays exactly the same.

Founders are still reviewing. Seniors are still making judgement calls. Quality still depends on a few people knowing what “good” looks like. AI speeds up activity, but it doesn’t change responsibility, ownership, or accountability.

This exact pattern came up repeatedly in a recent webinar I hosted on Leading AI adoption inside agencies, where I was joined by Zoë Blogg, Tom Davenport, and Andi Wilkinson. Despite running very different agencies, all three described the same reality: AI was being used in pockets, but delivery pressure, founder involvement, and margin challenges remained largely unchanged.

That wasn’t a tooling issue. It was an operating model issue.

The real problem is placement, not adoption

AI becomes genuinely useful when it is placed deliberately inside repeatable processes, not layered on top of messy ones.

When AI is introduced without clear workflows, decision points, and standards, it accelerates inconsistency rather than reducing workload. Outputs arrive faster, but trust drops, senior review increases, and founders stay involved because they have to.

As we discussed in the webinar, AI doesn’t fix unclear work. It exposes it faster.

Where AI actually improves agency profitability

Agencies that see real impact from AI are clear about which parts of their work genuinely require human judgement and which parts follow predictable patterns.

In practice, this means using AI for first-pass analysis and preparation so people are not starting from zero every time. It means drafting standardised outputs once “good” is clearly defined, rather than reinventing the same work repeatedly. It means surfacing knowledge, SOPs, and past decisions so founders and senior team members are not constantly interrupted to answer the same questions.

AI should support thinking, not replace understanding. That distinction matters, especially when quality and reputation are on the line.

This is where margin is usually lost — and where it can be recovered.

Why process design matters more than tools

A common mistake agencies make is introducing AI before their processes are clear. When workflows aren’t documented, decision logic lives in people’s heads, and quality standards are implicit rather than explicit, AI becomes risky rather than helpful.

The agencies seeing the best results focus on process first. They externalise thinking into SOPs, checklists, and standards, and then embed AI into those structures. This protects quality, reduces rework, and makes AI safe to use across the team rather than just by a few confident individuals.

Tool-led training doesn’t solve this. Concept-led training does. That was another strong theme from the webinar: understanding how to apply AI lasts far longer than learning a specific tool that will change again in six months.

The link between AI, founder time, and margin

One of the clearest indicators that AI is being used well inside an agency is whether founder involvement in delivery decreases.

If AI implementation still requires the founder to review everything, fix problems, or make final calls on routine work, then the underlying issue hasn’t been addressed. AI hasn’t failed — the structure around it has.

High-performing agencies use AI to remove the need for founders to be the glue holding everything together. That happens when decision logic is documented, standards are clear, and AI supports the repeatable parts of thinking that previously lived only in the founder’s head.

This doesn’t replace expertise. It protects it.

What to fix before adding more AI

For agency founders wondering why AI hasn’t delivered the promised gains, the answer is rarely to add more tools. The most effective starting points tend to be:

• Auditing where time is actually being spent across delivery

• Identifying which processes rely on repeated judgement rather than clear standards

• Documenting what “good” looks like before introducing automation

• Assigning ownership for how AI is used and measured inside the agency

These were the same conclusions that emerged during the AI adoption webinar, regardless of agency size or service mix.

AI doesn’t fix margins. Design does.

AI is not a silver bullet for agency profitability. It won’t compensate for unclear delivery, founder dependency, or under-designed operations.

What it can do — when used properly — is remove the repeated thinking, manual effort, and hidden work that quietly erodes margin over time.

For agency founders, the real opportunity isn’t “using AI more”. It’s designing an operating model where AI supports the business instead of sitting awkwardly on top of it.

That’s when margins start to improve.