# TMRE Insights Executive Retreat, Las Vegas 2025: themes of the day AI dominated the forms people filled in beforehand, but in the room the day turned to an older question: hiring, sponsoring and influencing so that the business still wants insight. Source: https://audiencestrategies.com/case-studies/tmre-retreat-2025-las-vegas.html Under the Chatham House rule: no names, no companies, no session-by-session account. The TMRE Executive Retreat met on Monday 27th October 2025 at Proprietor's Reserve, a private venue a short bus ride from the Las Vegas Strip, at the start of TMRE week. Roughly 45 senior insight leaders spent the day together: client-side heads of insight and analytics from many industries, with a handful of research and technology providers. The format was conversation first. Everyone opened by naming one honest struggle. The day then mixed on-stage conversations, short talks, a panel whose fourth chair rotated through the audience, an open-floor discussion of AI and a closing round in which the host passed the microphone from table to table. The day ran under the Chatham House rule: take any learning back to your team, but not the name or the company attached to it. These notes pool the day's thinking by theme, not by session. They carry no participant names, no participant companies and no sponsor names. Every position is framed as it was held in the room, by one person, by several or by one camp against another. The ideas are kept whole. The notes aim to carry no names or companies, and the host reviews them before publication. ## I. The worry underneath the AI worry: is the function still needed? The host's analysis of the forms filled in before the day found AI in nearly every response, often described as both the greatest opportunity and the greatest risk. But read closely, the forms carried a deeper anxiety. Respondents worried about proving value, about being absorbed by neighbouring functions, and about insight remaining optional rather than pivotal to decisions. They worried about data quality, about where the next generation learns the craft if machines take the groundwork, and about speed winning out over depth. The opening round said much the same in the first person. Some participants described an uncertain mix of excitement and weariness about AI. One described senior leaders arriving with the idea that research could simply be cut and replaced by AI. Another was working to prove the value of insight to a leadership team that places strong emphasis on direct revenue. One participant said their team's remit had expanded well beyond traditional research. Running through the round was the language of future-proofing. One leader asked how to future-proof a career that many younger people do not even know exists; others asked the same of their teams and their functions. Another asked how to stop long-term work on innovation and brand being squeezed out by the focus on growth today. Several were in the middle of merging teams: user research with market research, analytics with insight, or a scatter of groups into a new centre of excellence. The host closed the day by pushing on this from the outside. Finance, he argued, has had centuries to settle on how it is structured, and by and large it works; much the same is true of human resources. Insight teams are organised in endlessly creative ways, and on average they have less impact and respect than the function deserves. Perhaps, he wondered, there is too much flexibility, and the profession would be stronger if it settled on a right way. The room did not simply agree. The challenge was a response to one leader who had taken heart from the opposite reading: there is no single right answer, many different set-ups reach a good place, and there are bright people doing good work in all of them. That leader then turned the host's point around. The diversity of structures is not the cause of the impact problem, they suggested. It is everyone's attempt to solve it. ### So what for your team * Separate the AI question from the relevance question. A new tool will not settle whether the business thinks it needs you. * Name which neighbouring function could absorb yours, and make the case for what you do that it cannot, before someone else frames it for you. * When you restructure, be clear which impact problem the new structure is meant to solve. Structure is a means, not the answer. ## II. Hiring for character as well as credentials Several participants described a shift from hiring for credentials towards hiring for character. Curiosity was named most often, alongside coachability, a growth mindset, comfort with ambiguity, critical thinking and breadth. The working assumption, as those speakers stated it, was that most craft skills can be taught while the disposition to learn cannot. One participant argued for assessing aptitude beyond formal qualifications. Another described favouring people who can work across many data sources and connect the dots over deep specialists. Another noted that the hardest thing to find is someone who knows enough without arriving convinced they could already do the job, because that person is rarely coachable. Foundational craft still mattered. One leader wants people who can write a good survey question, and who know when a measure tracked for years has turned into research-speak rather than human speak and should be retired. Others flagged a gap in critical thinking among younger staff, who have had little chance to practise healthy scepticism and too often accept an answer, from a stakeholder or a machine, at face value. ### How to test for it Interviews are hard, because everyone has their best face on. Participants offered several techniques, each from their own practice: * Use interview panels of deliberately different personalities, and see who adapts. * Prefer a case study left with the candidate to situational questions, which candidates have learned to game. * Ask what they have taught themselves, and why. A course taken out of fascination says more than mandatory training completed because they were told to. * Ask whether and how they have used AI in research. The point is not a right answer. For the participant who suggested it, unbridled enthusiasm and outright fear are both warning signs; the goal is to hear how someone thinks. For growth mindset in an existing team, the advice was to show rather than tell. One leader redirects a team member who waits to be told what to do towards a peer who has already done the thing, so that they see the gap for themselves. ### The disagreements worth marking * Is the new generation the problem? One leader argued, deliberately against the grain, that younger workers are not wrong: a generation that refuses to give its whole life to an employer has seen the mistakes of the ones before, and is looking for balance, not idleness. Others agreed on the curiosity but worried about conflict avoidance and the habit of sending another message instead of having the conversation; they now coach people explicitly to set up time and have the conversation, even when it is uncomfortable. A third voice warned against generalising about any generation at all, recalling an earlier era when another cohort was dismissed as lazy and turned out to be outstanding. * Can AI help recruit? The host admitted that screening a stack of CVs against a job specification was one task he had never been able to make AI do well, and he ended up reading them by hand. One participant stressed the need for explicit evaluation criteria and human review when using AI. Leaders facing very large numbers of applications for a single role saw little alternative to some machine help. ### So what for your team * Write curiosity into the job description and test for it directly, through what candidates have taught themselves and why. * Keep a short list of non-negotiable foundations, such as writing a clean question, and teach them deliberately. * Coach the conversation, not just the deliverable. Show junior staff when a message should be a meeting. * If you use AI in screening, define the criteria first and keep a person checking the output. ## III. Backing people, not just advising them One distinction from the day was worth keeping simple: offering someone advice is different from actively backing their development. Advice is generous but costs the giver little. Backing someone means using your own standing to open doors for them, staying involved while they find their feet, and sharing responsibility if things go wrong. One participant resolved in the closing round to work out what that kind of backing should look like in practice, calling it a responsibility to bring up the next generation. A separate point came from the same closing round, about stakeholders. Before a senior colleague will use you as a thinking partner, you have to say, and then demonstrate, that what they share with you stays with you. The host admitted he had not thought about mentorship and sponsorship nearly specifically enough. ### So what for your team * For each person you lead, decide whether you are advising them or backing them. Backing means taking some of the risk. * Earn senior trust by showing that confidences are kept. ## IV. Influence at the top: from gatekeeper to gateway The host billed influence with senior leadership as the second most important topic on everyone's mind, after AI. Participants offered several answers rather than one. Participants described several ways to organise for influence, from specialist roles to people embedded deep inside business teams. One participant argued that structure matters less than it seems, because large organisations cycle through every model; what counts is the value added. One participant noted a cost of closeness: people embedded in a team find it harder to deliver bad news about that team's plans, so communication skills matter as much as research craft. ### Understanding the decision-maker One participant argued that being right is not enough: effective communication requires understanding the decision-maker's circumstances, not only the evidence. ### From validating to creating A recurring argument was that insight should stop acting as a gatekeeper and become a gateway: creating pull rather than pushing, and practising empathy for stakeholders as well as for consumers. One participant suggested asking why the business thinks it does not need insight, and what the function would have to change about itself. Another argued that a team that only says an idea failed should be asked for its next growth idea. Senior access is easier for evaluation work, should we do this or not, than for exploration, where one questioner argued insight adds most. The advice offered was to shift effort towards development gradually: pick a moment when the old way is plainly not delivering, make the change small and reversible, and prove it before scaling. One participant argued that research creates value only when people use it in decisions. ### The disagreement worth marking Should insight be made mandatory? A participant argued for building research into required steps of the innovation process, so the function cannot be bypassed. One reply was that required checkpoints have a place. Another reply was that compliance does not create demand; pull comes from bringing ideas the business wants. ### So what for your team * Decide whether influencing is everyone's job or a dedicated role, and recruit for it either way. * Before a senior meeting, ask what the decision-maker needs and fears, not only what the data says. * Pick one area and shift effort from evaluation to development, small and reversible first. * Ask where your work fails to reach the decision, and fix that before improving the work itself. ## V. Uncomfortable truths about the methods Two methods worries ran through the day. The first is fraud: as AI agents get better, telling people from bots in online surveys gets harder. The second is broader. One argument held that some long-trusted research methods deserve less faith than they get, and that the answer is to check what people report against independent evidence wherever possible rather than abandon the methods. Synthetic respondents drew a careful treatment. One argument was that the word covers very different techniques, and the profession needs a shared vocabulary before it can judge them. The case made for them was hybrid rather than replacement: validate against real people, and start where stakes are low, such as narrowing early ideas. One participant who had compared synthetic and real panels found the synthetic ones more rational than real people. The host noted that synthetic data barely came up in his interviews with insight leaders, he suspected because people were unsure whom to trust. Another participant challenged the usual picture of researchers lagging behind the technology: adoption should be judged on impact, not on keeping pace. A third argument was that when everyone has the same models, the advantage lies in data nobody else has, and in context: an AI tool that already knows something about the person it is analysing does better work than a clever prompt. ### So what for your team * Check your most relied-on methods against independent evidence before trusting them further. * Pilot synthetic data where stakes are low, validate it against people, and say exactly which kind you mean. * Treat your own first-party data as an asset to clean, connect and build on. ## VI. AI: the future is here, unevenly The day's open discussion of AI was framed by the host's interviews with about 40 insight leaders, some of them in the room, ahead of a talk later that week drawing on some 78 organisations. His headline borrowed a familiar line: "The future is here. It's just not evenly distributed." In his interviews, shining examples of AI benefit existed in every corner and the benefit in aggregate was large. Yet very few interviewees felt it had transformed their own team. Almost everyone he spoke to was struggling somewhere, which meant that honest sharing alone could multiply the benefit. ### Six links in a chain His diagnosis was six links, each of which can break the chain: * Executive sponsorship. Some leadership teams make vague proclamations about AI with no clue what it is for, and leave teams dangling. * Application approvals. Some people still have no good AI they are allowed to use with company data. * Foundational skills: good mental models of what AI is and is not good at, and the tips and muscle memory to get past errors. * Standardised tasks: playbooks, or shared assistants, for recurring work. * Process orchestration: joining tasks into end-to-end processes, which needs different technology, process engineering, and sometimes moving people and roles. * Sustained reinforcement. A burst of excitement followed by silence kills momentum. Almost nobody interviewed had all six. Then comes a choice about where the benefits go. Extraction takes them for the profit line: headcount back, or simply more output. Liberation shares them: more strategy and so-what thinking, adjacent work, time for innovation. Many leaders had not made the choice explicit, and for some it will be made for them. ### What the room added The discussion turned the frame into practice. * Guilt. One leader turned rough notes from two meetings into a structured research brief with a model's help. A senior colleague seized on a suggestion the leader would never have proposed alone, and the leader felt embarrassed. The host's response: we judge AI by standards we never apply to a conference, an agency or a report we borrow ideas from. * Blame. Insight teams often lead AI adoption. What happens when the next marketing disaster, a New Coke or a botched logo change, is blamed on AI insight? The host's answer was check, edit and own: a human must take responsibility for every output, so an AI mistake is a human using AI badly. A participant added that accountability should be joint, built into the process with stakeholders, never placed on insight alone. * Standards. Set guardrails for AI as you set them for survey design. One team is building a prompt library with defined quality criteria for sources and guidance on which model to use for what, and will communicate confidence the way it already does for a quick study versus a rigorous one. The host's caution: "a prompt is not enough." It needs preparation before, a process of checking and iteration after, and the proficiency of an expert to judge the result. * Privacy and intellectual property. One leader had seen a proposal to upload lists of employees to a consumer AI tool. The fix offered was enterprise licences with contractual guarantees, vetted by lawyers. * Stale sources. One participant warned that models lean on past online discussion and can confidently state a position that has since changed, correcting themselves only when challenged. The host's fix: insist on a search for anything timely, then click the source and read it. * The illusion of expertise. When anyone can sound expert, insight's new job may be to champion the integrity of the data and be clear where AI exploration stops and validated truth begins. * Plumbing. One participant described data-quality problems that required attention before further automation. * Confirmation bias. A tool built to let stakeholders query reports told users "you're absolutely right" when the data said otherwise. One check suggested: ask it to argue the opposite. Two signals pointed outward. One leader reported building agents on their own research data, so stakeholders query validated sources rather than the open web. Others noted that search is shifting from optimising for search engines to optimising for answer engines, and asked how buying through AI assistants will change the behavioural data insight teams rely on. On adoption, one participant argued that visible leadership support can make AI use feel more acceptable. The host's strongest recommendation: put a sceptical executive one-to-one with someone expert in AI, working on that executive's own job. Watching their eyes light up, he said, changes their mental model and their sponsorship. ### The disagreement worth marking Whose job is AI? In the closing round, one leader said it could not stay a side project; it needs someone whose real responsibility it is. Another said it is everyone's job, written into every person's expectations and commitments. The host's view was both: everyone, and a dedicated, usually overstretched champion to lobby for performance reviews, chase legal approvals and keep reminding people. One participant argued that sustained AI adoption requires accessible support and visible leadership. ### So what for your team * Score your team on the six links and work on the weakest. * Decide, and say out loud, whether the gains are for extraction or liberation. * Write standards for AI use as you would for research methods, and teach check, edit and own. * Get one sceptical senior leader a private, hands-on session on their own work. ## VII. The human edge, and the answer in the room The closing round returned again and again to the human side. One leader's takeaway was the importance of developing generalists. The host noted that in an age of AI you might have expected the opposite, a call for technical specialists. Others said storytelling was the skill that kept surfacing: insight people fall in love with the method, then waste the moment with a leader, which might be a corridor or a lunch queue rather than a meeting. The host recalled a brilliant mathematician, the opposite of a natural storyteller, who produced a compelling story every time an executive came near, and the effect it had on funding and priorities. The function's value, three participants said in different ways, is to bring emotion and meaning to data and to keep understanding people at the centre. One leader wanted to delegate more to AI precisely so that people could spend more time in rooms like this one, sharing problems. Another took comfort from the host's repeated point that the problems are hard but solvable. The host also put AI to the test. He had fed the day's transcripts to an assistant under an enterprise agreement and asked for the most useful lessons. The closing reflections had touched on one of its seven, the uneven spread of the future. The others, though several had come up earlier in the day, nobody had named in the closing round: move from gatekeeper to gateway; shift towards development over evaluation; leaders must model AI use themselves; practise stakeholder empathy, not just consumer empathy; proprietary data as the new advantage; and perfect is the enemy of progress. He doubted the point on proprietary data had come up as strongly as the summary implied, noted that AI loves a catchy phrase, and graded it at about 70 per cent: the rest he would cut. That was a lesson in itself: use the machine to remember, then check, edit and own. His closing message echoed the purpose he set out in the morning. For nearly every problem in the room, someone else in the room has already solved part of it. The value of the day lies in the connections that last beyond it. ### So what for your team * Develop generalists and storytellers as deliberately as technical specialists. * Use every moment with a leader, formal or not, to share one thing your team has learned. * Keep the network. The answer to your next problem probably sits with a peer. ## If there is only one page * The AI anxiety sat on top of an older one: whether the business thinks it needs insight at all. Know which question you are answering. * Those who spoke on talent look for character as well as credentials: curiosity, coachability, critical thinking and breadth, with a few foundations always, tested directly. * Backing someone is more than advising them. It means using your standing on their behalf and staying with them while they learn. * Influence is the job. One strong argument: be a gateway rather than a gatekeeper, and create pull rather than push. * Methods need honesty. Check trusted methods against independent evidence, and treat synthetic data as hybrid, validated and low stakes first. * The host found widespread examples of AI benefit in his interviews, but few reports of team-wide transformation. His diagnosis is six links: sponsorship, approved tools, skills, standard tasks, orchestrated processes and sustained reinforcement. Then choose extraction or liberation on purpose. * The future is here, unevenly spread. The answer is usually already in the room.