Why AI performance is a growth design question

There is a clear shift happening in how marketing is expected to operate. AI is moving from supporting individual tasks to executing workflows. Campaigns are now being planned, built and optimised with far less manual input and customer journeys are increasingly being shaped in real time. The expectation is no longer faster output — it is continuous execution.

Most leaders accept where this is heading. The harder question is what has to change underneath, so that AI translates into commercial performance rather than into more activity.

Why AI is present but impact is inconsistent

Most are already investing in AI. Pilots are running, tools are in place, and adoption is increasing across teams — so on the surface progress is visible. The commercial impact, however, tends to be inconsistent. In many cases AI is present across the organisation, but the effect on pipeline, conversion or growth is difficult to isolate, because activity has increased while the system that activity sits within has not fundamentally changed. That is where the gap tends to sit.

The leap from AI pilots to AI performance is less a tooling question and more a question of how the operating system around growth is designed. Agentic AI assumes a world where workflows are connected, data is usable, and execution can run with continuity — and in most marketing functions that environment does not yet exist.

Data tends to remain fragmented across systems, strategy is still largely static — captured in documents rather than embedded in execution — and campaigns are built and delivered as discrete efforts with manual coordination across teams. Ownership is often unclear, and the link between activity and commercial impact is not always well defined. AI does not resolve those conditions, it works within them, which is why many organisations are seeing progress in output without seeing progress in performance.

The four stall points where AI is present but impact remains inconsistent:

1. Tools get bought before the underlying data has been mapped and understood.
2. Automation gets built on top of data that is not consistent enough to act on.
3. Workflows look automated but do not improve from one cycle to the next.
4. The link between AI activity and commercial outcome is rarely defined clearly enough to know whether anything is working.

The three shifts behind real impact

What is starting to emerge is a different way of approaching the problem. The organisations we see making real impact are not treating AI as a layer to be added on top — they are using it as a reason to rethink how growth work is structured in the first place. That shift tends to show up in three ways.

Insight is becoming operational. Research, persona work and market intelligence are no longer treated as static assets — they are being turned into systems that can be accessed and applied in real time, at the point decisions are made. The value is no longer in producing the insight, but in how consistently it can be used.

Execution is becoming continuous. Rather than building campaigns one at a time, teams are creating workflows that can generate, adapt and improve outputs on an ongoing basis, so the unit of work shifts from the campaign to the system that produces it.

Growth is becoming a connected system. Marketing, sales and customer engagement are aligning around shared workflows rather than operating as separate activities, and AI is applied inside those workflows — where it can influence how opportunities are created, progressed and converted.

The conditions for AI to deliver growth

What we consistently see is that agentic AI only creates value when three conditions are in place. Data is consistent enough to act on, workflows are clear enough to execute without constant intervention, and human judgement is defined enough to guide decisions rather than react to them.

Without those conditions, AI tends to remain a productivity layer. With them, it becomes part of how growth is actually delivered.

The implication for most organisations is more practical than transformational. The starting point tends not to be scaling AI across everything at once, but taking one commercially important workflow and understanding how it actually runs today — where it slows, where it breaks, where ownership blurs, and where effort does not translate into impact.

We have seen this most recently in work with an international marketing function moving from individual AI productivity gains toward an agentic operating model. Rather than starting with the agent stack, the work began with two live campaigns — one in each of their core business lines — mapped end to end across data, process and execution. That mapping surfaced the gaps in the process and the gaps in the systems, before any automation was built on top of them.

From there, the work is to redesign the workflow so it can run with more consistency, more visibility and clearer decision points — and to apply AI inside that structure, where its impact can be measured and improved cycle by cycle. Each pass tends to reveal more, fix more and save more than the last, because improvements compound across campaigns rather than starting over each time.

What separates the organisations that are winning

AI will continue to develop, and the capabilities will continue to improve. In our experience, the organisations that benefit will be the ones that redesign how growth work gets done, so that AI has something meaningful to operate inside.

The question becomes less about how quickly to adopt, and more about how the system around growth needs to be redesigned to make that adoption count.