Cannes 2026: themes of the day

The retreat ran on Monday 22nd June 2026 in Cannes, as the executive retreat day of the Lions Insights & Strategy Summit. Roughly 45 senior insight and analytics leaders took part. It was invite-only and off the record, held under the Chatham House rule: anyone may share what was said, but not who said it. These notes pool the day's thinking by theme rather than by session. They carry no participant names, no participant companies and no account of who spoke when. Where a view belonged to one person it is reported as one person's view.

I. The day was billed as AI, and was about something older

AI dominated the pre-event forms and the opening round of introductions. The forms also carried a quieter ambition, to win a seat at the table and influence decisions rather than serve or validate them, which the spoken round largely drowned out.

The host's reading at the close was that very few of the day's conversations were really about the technology. They were about an older question of role and value, one the function held before AI arrived. AI magnifies that question, answers part of it, and is beside the point for another part. Few people were asking how to make a tool do a particular thing.

The bluntest version of the worry came in the introductions. One participant feared executives would cut the insights function and move the budget to AI to save cost. Another worried that the value of insight is quietly changing inside organisations. A third asked how insight drives growth at all when growth is hard to find across so many industries.

Not everyone was anxious. One participant argued the room was over-worried, and that AI belongs back in its place as one tool among many. Another was untroubled and simply curious about what people will do once agents take on more of the work. At least two said that heavy daily use had taken the fear out of it.

So what for your team

• Name the real question before reaching for a tool. If the worry is role and value, an application will not settle it.
• Treat the survival fear as a prompt, not a forecast. A function that can say what it is for, in growth terms, is in a stronger position than one that cannot.

II. Speed against understanding

One concern ran through the day. AI produces a plausible answer in seconds, and that speed can crowd out the slower work of establishing what is true. Faced with a confident answer today or a researched one in a few weeks, a business leader may well take the fast one. The narrower worry is that convincing output gets mistaken for reliable evidence.

One speaker put the question sharply: if the machine is more persuasive than we are, is it persuading people towards the truth, or merely towards an answer?

Several participants described the same pattern inside their own organisations: senior colleagues forwarding what an AI tool had told them, or arriving with a question they had "already checked with AI". One described the counterweight in their own business, where the instruction from the top down is to use AI and then apply judgement, so that nobody treats a machine summary of interviews as the finished job.

The picture was not one-directional. Some businesses hold two postures at once: data-driven parts lean in, while the parts that make the product or the brand stay cautious, leaving the insight team fielding opposite demands. In other organisations the insight team is the one pushing to bring AI in, against a wider business or a legal function that is wary of it.

A worry about the people coming next recurred, including in the host's opening. If AI both frames a problem and answers it, how do new entrants learn the craft of finding out what is true?

So what for your team

• Separate the two questions AI collapses: how fast can we answer, and are we answering the right problem? Give someone the job of checking the framing.
• Make validation the habit. Use the tool, then apply judgement and go back to the source.
• Plan deliberately for how junior people will learn the craft when the tool does the first draft.

III. Two answers to the function's future

The host observed at the close that many topics produced two opposing answers, both defensible. On the function's future he drew out two sets of conversations that pull against each other.

The first held that the function's job is to guard the truth. If insight teams are no longer the owners of knowledge, the job becomes curating clean, trusted data, removing the noise, keeping the signals fresh, and making sure that what is true reaches the point of decision, by AI or otherwise. That is a different skill set and mindset from owning answers.

The second held that data is not the point. Insight leaders are experienced and senior, and should own decisions, champion them and help make them. On this view the sooner the work moves from data up to judgement and storytelling, the better.

The host did not try to settle it. His point was that the function has real freedom to choose its own future, and an obligation to use it.

The forecasts offered were mixed. One participant expected some companies to fold insight into strategy and technology, others to keep it, and the function perhaps to re-emerge under another name. Another reported seeing it downsized, and sometimes cut, at some firms. A third worried it could be absorbed into strategy, finance or technology because its work is seen as mining data rather than understanding people.

Two challenges complicated both answers. One participant asked why the integrating role should belong to insight at all, rather than to research and development, technology or marketing. The same participant argued that empathy, long claimed as the edge a machine cannot reach, is a weakening defence, because software already learns individual patterns and builds apparent rapport.

So what for your team

• Choose your stance on purpose. Guarding the truth and owning the decision ask for different skills; drift picks badly.
• Do not rest the whole argument on empathy. One participant questioned whether it remains a sufficient defence.
• If you want the integrating role, earn it on the merits rather than claim it.

IV. Influence is the job, and it is a craft

A wide thread argued that a finding is not the end point. One participant argued that research should be judged by its practical effect on the business. Influence was described as resting on listening to scattered needs across an organisation and on trust: findings can be imposed, but people will not act on them without it.

One participant argued that the method of influence should change with the type of decision, with storytelling counting for more the more senior and strategic the decision.

One participant argued that faster research is useful only if it improves the eventual business decision.

Views differed on whether to frame work as a threat or an opportunity. One participant counted failing to establish what is at stake among their own biggest mistakes: without stakes there is no tension, and nobody cares. A different view held that a problem need not be existential, only an unanswered question, and that the opportunity follows once people accept there is something they do not know. A separate point was that purely confirmatory research is hard to justify now.

Other practical points from the day:

• Involve decision-makers early and agree how the findings will be used.
• Prioritise research that stakeholders can absorb and use.
• A thing barely exists in an organisation until it has a name people repeat.
• AI could hold the institutional knowledge that leaves whenever people move roles every few years.
• Always answer "so what?" for the audience in front of you.

So what for your team

• Match the method to the decision rather than using one register for every audience.
• Involve stakeholders early, not only when the results arrive.
• Build trust before the message has to land, not at the moment it does.

V. The work itself is changing

Several participants noted that AI assistants increasingly sit between brands and buyers, and that brands may now have to satisfy the tool that mediates a purchase as well as the shopper. The practical consequence is to test how automated recommendations represent your products.

Automation also raises the demand for evidence. If more decisions are made or prepared by software, that software needs timely, reliable inputs, and people need to be responsible for their quality and use. Several voices saw that as a natural role for insight teams.

Organisational barriers were a recurring concern. Panellists and guests described insight and analytics spread across separate teams, budgets and reporting lines, each holding a slice of the picture. One added that easy AI storytelling tools make this worse, because anyone can now produce a polished narrative and competing versions of the truth multiply.

Two participants disagreed about whether insights and analytics could merge. One argued there are fundamental reasons they cannot; the other expected AI to force the merger, and soon.

So what for your team

• Find out how the main AI assistants describe your products and categories.
• Decide who owns the quality of consumer data end to end, before someone else takes the seat.
• Add commercial fluency to the data and the empathy you may already have.

VI. Making adoption useful

The table discussions turned to what makes AI adoption work. One table first questioned the premise that it is failing. Their answer was yes and no: it disappoints against a leadership expectation of near-miraculous efficiency, yet use keeps spreading.

That table's main conclusion was a reframe. Adoption seemed to work where the starting point was growing the business, with efficiency following, and to struggle where AI was treated as an efficiency exercise. They asked whether organisations are measuring the right things. Members argued for working backwards from the outcome, such as revenue, share or brand equity, and for redesigning the process around what the technology can now do, rather than slotting an agent into one step of today's workflow.

Other tables reported a mix of successes and pitfalls:

• Start small and scale; consider running insight as a source of revenue rather than a cost.
• Agree rules on what data may go into which tools.
• Pitfalls: functions each running their own AI effort without coordination, and buying licences without training or any plan for adoption.

Several tables traced failures to organisation rather than technology: useful outputs that were never built into anyone's working week, people resisting out of fear of being replaced, and internal technology and licensing rules, often justified on security grounds, that stop teams using their own data. One participant observed that most use is still simple chat, while the people getting real value are those who already code and work with data.

On cost, participants reported substantial spending on AI usage and executives starting to ask what the bill buys. One participant's view was that nobody has worked the economics out yet. One participant questioned whether usage measures reliably indicate business value.

There were speculative exchanges too. One participant was sure a one-person, very high-value company is coming, and thought agents could in principle run a business. The limit raised was whether models stay current as consumers change, which is why two participants said they distrust synthetic data: it may hold for a while, then behaviour moves.

So what for your team

• Start from the outcome you want to grow and work backwards, rather than adding AI to an old process.
• Build time and ownership for using new outputs into routine work.
• Investigate workflow and organisational barriers alongside technical performance.
• Review performance measures against the outcomes they are meant to support.
• Keep evidence and limitations attached to findings when they are reused.
• Hire for curiosity, adaptability and reasoning, and build a team of complementary skills rather than similar people. One participant now treats everyday AI use as a baseline requirement for new hires.

VII. The human edge, and the choice

The host noted that two or three people, on stage and in private, had said versions of the same line: if a machine can do a task, that cannot be where a person's or a team's value sits. People admitted candidly that tasks they, or their senior colleagues, used to rely on them for can now be done by a machine. He also said the examples of advanced AI work he heard were further along than he expected, and that with capability accelerating, the implications are near-term.

One participant argued that the human still owns the decision. Even where one AI tool serves many brands or markets, each decision has to fit its own market, customers and economics, and in their view that is still human work. A table reported a related point: raw data work may shrink, but interpreting it for senior leaders needs a trusted human voice in the room, not least because AI can be confidently wrong, and a polished wrong answer can be well received. Their advice was to be willing to say the unpopular thing.

Two warnings came from one table. People reach for a reassuring outside name to feel confident, even when that adviser works from the same data everyone already has. And organisations may be training people only to operate the tool at a surface level, rather than preparing them for what they will contribute once the tool absorbs that layer of skill.

The host's closing theme was choice. The technology may be inevitable by now; its impact in an organisation is not. How it is set up and guided, which uses it is pointed at, and how people are helped to use it or to reskill are all choices. He widened it to the next generation and society: do we automate all work that can be automated, or keep people in the loop even where they are only a little helpful?

He returned to a line from his opening. The future is already in the room and in most of the organisations represented; it is just not spread around nearly enough. For many problems one team is wrestling with, another has already found part of the answer.

So what for your team

• Take the value question seriously: move time and hiring towards what only a person can do.
• Keep a trusted human in the loop on the decisions that matter.
• Be deliberate about the future you want; the technology may be set, but its use is a choice.
• Spread the capability that already exists inside your organisation.

If there is only one page

• Billed as AI, the day was really about role and value, an older question the technology magnifies.
• AI buys speed and risks understanding. Separate how fast you can answer from whether you are answering the right problem, and make validation a habit.
• Two futures were defended: guard the truth, or own the decision. Choose on purpose.
• Influence is the job and a craft: match the method to the decision, involve people early, and build trust before the message lands.
• Several participants described organisational barriers to putting AI outputs into use, and automation raises the demand for reliable evidence.
• Adoption worked better, in one table's view, when it started from growth rather than efficiency, and it stalled on organisation, incentives and habits as much as on tools.
• One view placed the human edge in owning the decision and saying the unpopular thing. The host's closing reading was that the rest is a choice, and much of the answer is already in the room.