By Virgine Goupilleau, Founding Partner of BerylliumIV
Here is a number worth sitting with before Cannes. Of the workers now saving more than a full working day every week through AI, 66% are receiving no guidance on what to do with that time. In an agency, that time does not sit idle. It gets absorbed straight back into client servicing at existing rates. No margin increase. No record of where the productivity went. No conversation about what the organisation should have done differently with it. The gain dissolves into the working week without commercial trace.
The technology is delivering exactly what it promised. The failure sits in the operating model. And it is the defining commercial challenge for every agency that has spent the last two years building an adoption story without stopping to ask what comes next.
The pattern is familiar. A new capability arrives. The industry embraces it faster than most. Adoption gets tracked and celebrated. And then, quietly, the organisation carries on doing things almost exactly as it always did, with the new capability layered on top rather than structurally embedded. AI is the most consequential version of that pattern the industry has ever faced. BCG’s *AI at Work 2026* report, published this month based on responses from 12,000 respondents across 14 markets, has now precisely measured where it is repeating.
The headline finding is striking. 74% of frontline workers are now regular AI users, up 23 percentage points in a single year. The adoption debate is over. What has not been settled is what comes next.
The question the industry is not asking
There is a number in the BCG report that should change every agency budget conversation for the next twelve months:
- Strategic clarity around AI lifts measurable business impact by 25 percentage points.
- Better tools, without strategic clarity, lift it by 5.
Five times the return! Not from a better platform that everyone has access to, but from the organisational architecture that determines whether the value of those tools ever reaches the P&L.
We all know how AI investment decisions actually get made. Someone in technology or operations makes the case for a new tool. The CFO signs off the licence. Adoption gets tracked. The story ends there. What nobody measures, because it is harder to quantify and harder to sell internally, is whether the organisation has been redesigned to receive the value the tool creates. Licence costs are visible on the P&L. The operating model deficit is not. The gap stays invisible until it shows up in a margin conversation that nobody quite expected.
BCG’s data is unambiguous. Without that redesign, you are capturing one-fifth of the available return. The other four-fifths are sitting in the chasm between having access to AI and having an organisation built to use it.
Independent research published in March 2026 by INSEAD and Harvard Business School puts empirical weight behind exactly that finding. A field experiment tracking 515 high-growth firms through an AI adoption programme found that firms simply given AI tools saw marginal gains. Those coached to rethink where AI creates value across their production process saw something fundamentally different: 12% more tasks completed, 1.9 times higher revenue, and 18% greater likelihood of acquiring paying customers. The researchers named what separated them “the mapping problem”: discovering where and how AI genuinely creates value within a specific firm’s production process. The firms that solved it did not use AI to do the same things faster. They used it to do different things entirely.
That distinction matters enormously for agencies. When AI generates time, redeployment is straightforward. What most organisations lack is anyone with the authority and diagnostic framework to decide where that time actually goes. The time gets absorbed, the cognitive load compounds, and the organisation ends up running faster without running differently. That is the efficiency trap: a technology built for transformation being used to accelerate a model that has not changed. The industry is at serious risk of optimising itself into the wrong future.
What transformation actually requires for our industry
This is where BCG’s report reaches the limit of what a global survey can say, and where industry-specific experience becomes the more useful lens.
An intelligence operating model, the frame that Cannes Lions has placed at the centre of this year’s AI conversation, demands specific organisational work, with identifiable components, most of which are unglamorous and none of which get discussed in festival sessions.
Auditing where the time is going. AI-generated savings do not automatically become strategic capacity. Someone has to audit where they are currently going, at workflow level, and make an active decision about where they should go instead. That conversation does not happen naturally. It requires systems thinking, operational authority, and a diagnostic framework to run it.
Building a gate system for experiments. 58% of organisations in BCG’s survey remain in what the report calls “replace” mode: substituting AI for human tasks one-for-one, without the process redesign that would allow AI to generate genuinely different outcomes. The reason is almost never technical. Nobody has built the framework for deciding what an experiment needs to demonstrate before it moves to production, or what happens when it does not.
Establishing governance for agents already in production. BCG found that agent integration doubled in a single year, from 13% to 30% of organisations. Simultaneously, 50% of those organisations have no clear governance framework for what those agents are authorised to do or who is accountable for them. To make that concrete: an AI agent operating within a client’s media-buying workflow, with no defined authorisation boundaries and no named accountability, represents an unsanctioned liability. If it makes a decision that causes a brand safety incident or a compliance failure, the question of who is responsible does not currently have a clean answer in most agencies. That liability exists right now, recognised but unaddressed, in most agencies.
Reading the talent signal correctly. 67% of regular AI users say AI has improved their job satisfaction. 41% say it has increased their cognitive load. Both things are simultaneously true, and the reason is straightforward. When AI arrives without redesigning the work around it, people manage AI on top of their existing responsibilities rather than through restructured ones. Early novelty carries the satisfaction score. Then enthusiasm fades, the load compounds, and what follows is attrition, specifically among the people most capable of working effectively with AI and who have the most options about where they do it. Only 28% of BCG’s respondents see a connection between what their leaders say about AI and what their organisation actually does. For an industry heading to Cannes to talk about the intelligence operating model, that number should give pause.
From adoption to transformation: the work that actually needs doing
The intelligence operating model is the correct frame. The shift from organising around media placement to organising around intelligence, where AI operates across creative, production, media, strategy, and workflow simultaneously, defines what a competitive agency looks like in three to five years.
But naming the destination is not the same as building the road.
The INSEAD and Harvard research is instructive here too. The firms that shifted AI use toward production redesign and strategy, rather than task automation, saw revenue 1.9 times higher than the control group. Their demand for external capital fell by 39.5%, because they were building businesses with genuine structural advantage rather than marginal efficiency gains. The same dynamic is available to agencies. Not by copying what startups do, but by understanding what the research reveals about where AI actually creates value: not in completing tasks faster, but in rethinking which tasks the organisation should be doing at all.
The agencies that emerge from this period with a structural advantage will be the ones that made the uncomfortable decisions now: Where does value actually get captured, and who is accountable for it? What does the workflow look like when it is genuinely redesigned rather than augmented? What governance is in place around AI already making decisions in client-facing processes? When AI generates capacity, what strategic purpose does the organisation redirect it toward?
Underpinning all of it is a question of enterprise design: whether the AI capability stack has been deliberately architected to compound into competitive advantage, or simply accumulated tool by tool until someone notices the margin has not moved.
Those decisions require a different kind of leadership conversation than the industry is currently having. Not about which tools to buy. About what the organisation needs to become, and what specifically has to change for it to get there. What that conversation requires, increasingly, is a new kind of role inside the agency: someone who sits at the intersection of AI capability, commercial model design, and workflow architecture. Not a technologist and not a traditional strategist. The M-shaped orchestrator, fluent across disciplines and accountable for how the intelligence operating model actually functions in practice, is the organisational answer to the mapping problem INSEAD identified. No job description for this role currently exists at scale. That is itself a signal of how much structural work remains.
BCG has done the measuring. The productivity gains are real and being wasted at scale. The strategic clarity that would convert them into business value is absent in most organisations surveyed. Governance for agents in production does not exist for half of them. And the gap widens precisely where adoption is highest.
Cannes is a good place to name these things. The conversations about the intelligence operating model happening in the Palais next week are overdue and genuinely important. But the work that follows those conversations is not a festival conversation. It is an organisational decision, made in a boardroom, about whether agencies will redesign themselves around what AI actually makes possible, or keep layering new capabilities onto an unchanged structure and call the result a transformation.
Every week that gap stays open is another week the productivity gains dissolve into margin that nobody captured.
The industry has spent two years building the adoption story. That part is done. The transformation story is the one that actually needs to be written, and that’s the most exciting one.
