Updates to the panels and the applications, and the studies we read while building them.
The panel can now also be drawn from respondents interviewed in 2025–2026, in the countries where we have them. When you order you choose the period of the profiles: the older ones stay included in the price, the recent ones are a €39 add-on. For any order saved up to and including 31 October 2026, the add-on is free. Recent profiles matter most for questions about prices and spending.
Every panel template now shows you the questions we put to people. We have already chosen them: you can untick a question, change its wording or add one of your own.
For products that do not have a price yet. People see the page without prices and say for themselves where it feels cheap, where it feels costly and where they would no longer take it. You get the range they settle in, beside that of an ordinary product.
To tell whether a result is good or just average, we ask the same questions about something ordinary in the same category. The field now comes filled in by us; you change it only if you have a specific competitor in mind.
SKPTIK panel runs now come with a short film of the panel, alongside the report and the PDF. You open it from your account, and the PDF prints its address as a link and a QR code, for whoever reads the report on paper.
Synthetic respondents capture the broad trends and the order of brand preferences, but overstate positive attitudes and vary less than real people. In the study they matched the real choice about 79% of the time and were off by 1.2 points on average on a 7-point scale. The authors recommend them for early-stage concept tests, not for decisions that need precise figures.
Read the articlePyMC Labs and Colgate-Palmolive compared synthetic answers with 9,300 human responses from 57 surveys on personal care products. When the model answers in words and the rating is derived from the text afterwards, it reached 90% of human test-retest reliability. The result concerns concept tests in well-defined contexts.
Read the articleA public dataset of 2,058 people in the United States, each with answers to more than 500 questions, collected in four waves. It is built for developing and checking digital twins. The last wave repeats earlier tasks, so it shows how stable the same people’s answers are.
Read the article