Reliability · ESOMAR
ESOMAR's 20 questions. Answered.
ESOMAR, the international association for market research, proposes twenty questions for buyers of AI-based research services. These are our answers. The questions are summarised; the figures are those of the reading of 12 October 2026.
Company profile
1 What experience and know-how does your company have in AI-based solutions for research?
Valyron is a young company, based in Madrid, that does only this: simulate how a population would respond to a product, a price or a message before it is launched. The service is in beta today, with access by invitation, in 21 European countries.
The method builds on the published research on simulated populations, and each country is measured against real answers before it is offered. The figures, with their date, are on the reliability page.
2 Where can AI improve research? What does it bring and which problems does it address?
In the early stage of a decision: choosing between versions of a product, finding the price people do not find expensive, understanding the reasons for and against, and seeing which audience it suits best. A simulation gives that reading in minutes, so more ideas can be tested before investing.
It does not replace research with real people when a market measurement is needed: it gets you there with fewer, better chosen options.
3 What practical problems have you met in using and deploying AI? What has worked well, what less well, and why?
What works best is comparing: which version wins and why. Running the same study five times, the same version won every time.
What works less well, and is therefore published as a limit: absolute levels are indicative, differences between audiences come out softer than in reality, and price is read better through what is perceived as expensive than through purchase intent.
Is it explainable and fit for purpose?
4 Can you explain the role of AI in your service in simple terms? What are its key functions?
You describe your product (up to three versions, or several prices of one version), choose where to ask and which audience, and add your own questions if you wish. Hundreds of simulated people, each with their own profile, read it and answer as that person would: whether they would buy it, which one they prefer, what they think of the price and why.
Then the analysis groups their reasons, compares versions and audiences and writes a recommendation. You get a report with verbatim quotes, Excel, PDF and a read-only link for your team.
5 Which AI model do you use? Do you build it yourselves or integrate an existing system from another company?
The answers and the analysis are generated by language models from providers bound by a data processing agreement, which we call through their cloud service. We do not publish which models or versions: that is part of our know-how, like the rest of the method.
What is Valyron's own is everything around the model: how each population is built, how the questions are asked, how the answers are aggregated and how we check that they resemble the real ones. Who processes data and for what is in the privacy policy.
6 How do the algorithms deliver the intended result? Can you summarise the underlying data and how it combines with the model?
Each simulated person is a profile built from population statistics of their country, with their age, sex, education, region and opinions. Each country's population reproduces the make-up of its society and is the same in every study, so two studies can be compared.
The model receives the profile and the concept and answers as that person would; the answers are then aggregated in the report. The statistical sources, with the attribution their licences require, are on the sources page.
Is it trustworthy, ethical and transparent?
7 How do you verify and validate the output? How do you measure validity and spot unreliable or biased results?
We compare what the simulated population answers with what the real population answers to the same questions. In Spain, the reference country, its answers come 28% closer to real people's than those of someone filling in a questionnaire without paying attention; and running the same study five times, the same version won every time.
Each country passes that comparison before it is offered, and every change to the method passes it again before it reaches the service: if it gets worse, it is not adopted. Figures based on a small sample are flagged in the report as indicative.
8 What are the limitations of your models and how do you mitigate them?
We publish them: it is a simulation and the answers are not from real people; it is for comparing versions and audiences, not for forecasting sales figures; absolute levels are indicative; differences between audiences come out softer than in reality; and price is read better through what is perceived as expensive.
We mitigate them in the report itself: it leads with the comparison between versions, which is what holds, flags cells with a small sample and gives every figure its reading. The full list, with its date, is on the reliability page.
9 Have you designed the service with a duty of care to humans in mind?
No real person answers our studies: nobody is surveyed and nobody's data is collected, and no simulated person corresponds to anyone identifiable. To run a study we ask for no personal data: only the product, its versions or prices, the audience and your questions.
Towards whoever decides, the report always says it is a simulation, that it guides and does not replace a market measurement, and it promises no accuracy we have not measured.
How is it overseen?
10 Transparency: how do you make it clear when AI is used in any part of the service?
Everywhere: the website, the form and every report say that the answers are simulated, not from real people, and the report carries the notice «Valyron simulation. It guides decisions; it does not replace a market measurement.». The quotes in a report are texts from the simulated people, not real testimonials.
11 Do you have ethical principles defined for the solution? How do they shape its behaviour?
Yes, and they work as product rules: we ask for no personal data to run a study; we call the people «simulated» and never present them as real; we publish the limits next to the figures; we promise no accuracy we have not measured; and your concepts are used only for your study.
The rules that affect what is shown (the wording, the limits, where each figure comes from) have automatic checks that stop a version of the website that breaks them from being published.
12 Responsible innovation: how does your solution build in human oversight?
People decide at both ends. At ours, every change to the method is decided by a person on the team and is not adopted without being measured against real answers again. At yours, the report recommends but the decision is yours: that is why it shows the reasons, the quotes and the limits next to the recommendation.
If a study fails or ends with problems, the report says so and it does not use up your quota.
Data governance
13 Data quality: how do you check that the underlying data is accurate, complete and relevant to the objective?
The data the result depends on is that of the simulated population. Each population is built from statistics of its country and, before it is offered, we check that it reproduces the make-up of its society and that its answers come close to those of its real population. An approved population is frozen: it does not change without being measured again.
14 Data lineage: do you document the origin and processing of input data? Are those sources available?
Yes. The statistical sources of each population, with their edition and the attribution their licences require, are published on the sources page. Each population is an identified version, and each report is tied to the version of the population and of the method it was made with.
15 Link to your privacy notice.
It is at valy-ron.com/en/legal/privacy (in Spanish, valy-ron.com/legal/privacidad) and it is the same for the whole service.
16 What do you do to comply with data protection law and protect participants' privacy? Do you have a legal basis for the processing?
There are no real participants in our studies: simulated people are not people and do not correspond to anyone identifiable. The personal data we process is that of your account, with the legal basis, purpose and retention period set out in the privacy policy, under the GDPR.
To generate the answers, the text of your concept is sent to language models from providers bound by a data processing agreement, never your identity. You can download your data or delete your account from «My account», or write to us at contact@valy-ron.com.
17 How do you make the system resilient to attacks, noise and other disruptions? Which security frameworks and standards do you follow?
All traffic is encrypted over HTTPS and the browser is required to use it; each account sees only its own studies, and the database enforces it with row-level access rules; the service keys never leave the server; shared links are read-only and can be revoked at once; and we rate-limit requests to curb abuse. A study that cannot complete is closed with a notice and does not use up quota.
We do not hold ISO certifications today.
18 Data ownership: do you define and communicate data ownership, intellectual property and usage permissions?
Yes, in the terms of contract: the concepts you give us and the reports you get are yours, and you authorise us to process them only to provide the service. The software, the method and the Valyron brand are ours.
19 Data sovereignty: do you restrict what can be done with the data?
Yes. Your concepts and reports are used only for your study and are seen only by whoever signs in with your account or receives a link you create and can revoke. We do not sell or share them. Your account, concepts and reports are stored in London (United Kingdom).
The only exception is voluntary: if you record in a report what happened when you launched the product and tick the box that allows it, which starts unticked, we may use that data in aggregate, anonymous form to improve our predictions. You can withdraw that permission at any time.
20 Output ownership: is it clear who owns the output?
Yes: the report is yours and you can use and share it inside or outside your organisation. When quoting it, we recommend saying that it is a simulation.
If you need these answers signed or any further detail for your procurement process, write to us at contact@valy-ron.com.
Validate your next idea before you launch it.
Results in minutes, at a fraction of the cost of traditional research.