The FlashInsight feed.

We will be at Printemps des Études in Paris, at the Palais des Congrès, Porte Maillot, on 24 and 25 September 2026. Come and find us on the stand — and feel free to argue with us.

The workshop. On 24 September, 4pm to 5pm, room 352A, Laurent Florès chairs “Synthetic research: four field reports,” part of the Expertise Benchmark sessions. No product demo. Four practitioners talking about what they actually did, what worked, and what did not.

Joining him: Florence Hussenot, CEO of Adwise · Guillaume Valicon, co-founder of Airpanel · Thibault Geenen, France CEO of Unguess · Bruno Walther, CEO of Imagine all the people.

Why that format. AI was everywhere at the ESOMAR congress in Valencia this month, but its most visible use is not the one you would expect. It is the AI-assisted interview, where a model runs the conversation and the respondents are entirely real. That changes the arithmetic of qualitative work: instead of twenty interviews you can run several hundred. Qualitative is scaling up.

The funding tells the story better than the talks did. Conveo announced a $50M round on 2 September, while the congress was under way. Listen Labs puts its total raised at $100M. Verso and others are working the same ground. Those rounds signal investor appetite, and a market where advertisers are genuinely starting to commit.

Synthetic research came up in every conversation — synthetic personas, digital twins, synthetic data. Plenty of debate, and plenty of confusion with it. The same words get used for different things, and the promises usually run ahead of the evidence.

Buyers are neither enthusiastic nor dismissive. They want three specific things: be clear about what each use is for, say how the method works, and show enough of it that they can assess the calibration for themselves. Our profession has asked for that discipline for years. On this subject, buyers are not getting it.

So we publish. Our answers to ESOMAR's twenty questions, including the ones we cannot yet answer. Our validation work. And our insistence on keeping these definitions apart instead of running them together.

The ESOMAR Congress opens tomorrow, 2–3 September. We will be there. We wanted our account on the record beforehand: written down, with the figures and the method needed to judge it.

Our answers to the twenty questions. ESOMAR publishes twenty questions a buyer should put to any supplier of AI-based research. We have answered them publicly, one by one, including on what our method cannot yet do. Re-reading ourselves, we found three claims on our own pages that did not hold: we have corrected them, and the position note says which.

What a buyer can check. A position note argues that calibrating a synthetic panel does not require owning a private panel, and that source data a buyer can verify is worth more than performance a buyer must take on trust.

The number the market keeps getting wrong. “85%” has become the industry's shorthand for “synthetic respondents work”. It is not 85% accuracy: it is 85% of the consistency the same people manage with themselves when re-interviewed. That distinction changes what you can reasonably buy.

A test anyone can run. The last document sets out a simple method for measuring whether simulated respondents are too alike, by comparing them against the same people re-interviewed — which is what gives the number a scale. It works from public data and costs nothing. We apply it to our own panel as well as to others, and we publish the results it produces. It also carries what we committed, in our answer to question 5, to make public either way: the re-analysis of a domain fine-tuned model trained by another team, which did not persuade us to adopt one.

Method over Magic.

François Bazini and Dr Laurent Florès have published an op-ed in Adweek: AI now writes flawless-looking marketing plans — and that is exactly the problem.

Who François Bazini is. A global brand leader, French- and US-educated, former BCG strategy consultant, he has held marketing roles at Danone Canada, PepsiCo (New York and London) and Suntory, leading international brand portfolios in spirits and beverages. He writes on scaling brands across countries and cultures.

The argument. A frontier model will draft a full plan in minutes: market summary, SWOT, recommendations. The tone is confident, the format is impeccable. Weak thinking used to look clumsy; now it arrives beautifully formatted.

Four traps. Summary is not synthesis — the model describes a competitive landscape but never settles the “so what?” for one brand in one context. Novelty bias — trained on what the internet celebrates, it leans toward launches and extensions rather than focus, producing motion without strategic direction. Surface appeal — exciting activations that quietly contradict the brand's economics. Context-blind imitation — importing a competitor's playbook without the conditions that made it work.

The hybrid CMO. The answer isn't to refuse AI, but to couple silicon with human judgment: AI brings speed, reach and analytical power; experience brings the trade-off discipline and the scar tissue — failed launches, extensions that add complexity instead of growth, elegant plans mistaken for strategy. In practice: let the model draft, interrogate it on its logic and assumptions, then make the call yourself.

Why this is our fight. What we do at FlashInsight isn't about AI for its own sake: it's about marketing decisions. We are convinced AI genuinely helps — and that the future is neither all-AI nor all-human. It is hybrid: a hybrid CMO, just as there is now a hybrid researcher.

Read the full piece in Adweek — link below.

An independent first validation confirms the robustness of FlashInsight's population-based sampling method. More waves — and a European edition built on open data — are coming.

Paris, June 30, 2026 — FlashInsight — synthetic market-research solutions developed by Dr Laurent Florès — today released the results of a first validation of its US panel against Twin-2K-500, a public dataset of 2,058 real Americans published by Columbia Business School (Marketing Science, 2025). FlashInsight's synthetic respondents — built only from public population statistics — reproduce real American consumer responses as faithfully as Columbia's individual “digital twins.”

What it means, simply. People never answer exactly the same way twice; that level of self-consistency sets the limit any prediction can reach. FlashInsight reaches the same level as Columbia's twins (a 0.83 correlation) — all the more striking because Columbia's twins are built from each person's real answers, whereas FlashInsight uses only public data and replicates no individual.

Beyond the numbers. The synthetic respondents don't just match averages: they are psychologically coherent — their attitudes follow the same personality and values patterns found in real people. The method is sound, not a fluke.

What's next. This first wave will be followed by others across more domains. And because the method needs only open data, FlashInsight is preparing a European edition built on Europe's rich open data — without digital twins, which require private individual data. Researchers, brands and partners interested in taking part in the next waves are welcome.

Validated for aggregate audience research; a complement to, not a substitute for, real fieldwork on a specific creative.

Academic reference: Toubia, O., Gui, G. Z., Peng, T., Merlau, D. J., Li, A. & Chen, H. (2025). Twin-2K-500: A Dataset for Building Digital Twins of 2,000 People. Marketing Science. arXiv:2505.17479 (CC BY 4.0).

FlashInsight has teamed up with Insights House, a Casablanca-based research institute known for its sharp read on Moroccan and African consumers, to bring synthetic-persona pre-testing to the Moroccan market.

The method is the one we run everywhere — calibrated synthetic personas, held to a rigorous protocol — but here it's built around the Moroccan consumer. Calibration starts from official statistics (the HCP census), layers in institutional sources (Bank Al-Maghrib, the ANRT, the CESE), and draws on a proprietary 2026 study by Insights House into how Moroccans relate to brands, advertising, tradition, modernity, and “Moroccanity.” That study makes the personas sharper still.

The full toolkit is live — quantitative and qualitative alike (one-on-one interviews and synthetic focus groups). And the first clients are already on board: advertising agencies, plus local and international advertisers.

On the ground, the partnership is led by Ainad Bachir (former VP at Ipsos Canada, former CEO of Nielsen Maghreb, co-founder of OpinionWay Maghreb). It's FlashInsight's playbook in action: one method, adapted market by market, with the partners who carry its rigor on the ground. Method over Magic.