What Preferences AI Research is
Each Market Insight answers one market question for Asia-Pacific (for example, how Southeast Asian shoppers pay online) with a set of individually citable statistics. Every statistic shows its source, its date, and an evidence level, and has its own link so writers can point readers to the exact figure.
Market Insights combine two kinds of evidence: published figures from governments, statistics offices, and original reports, and outputs from Preferences AI Digital Population simulations. The two are always labelled differently.
Evidence levels
- Primary sourcePublished by the organisation that collected the data: a government department, statistics office, central bank, regulator, or the original author of a study. We link to the original publication.
- Secondary sourceReported by a third party that did not collect the underlying data, such as an industry summary. We use these only when the original is unavailable and say so.
- PAI simulation · directionalOutput of a Preferences AI Digital Population simulation. It indicates the likely direction and relative size of a preference in the simulated population. It is not a census count or a survey of real people.
How simulations work
A Digital Population is a synthetic panel calibrated on public census and statistical-agency data and on social listening for a specific market. When we run a Quick Test, Scenario, or Survey, each synthetic respondent answers according to its modelled traits, and we report the aggregate distribution.
Simulations are useful for comparing options (which message, which price band, which channel) and for spotting likely differences between segments before running fieldwork. They are not a substitute for measured behaviour.
Synthetic data labelling policy
For every simulation statistic we publish:
- It carries the PAI simulation · directional badge and a link to this page.
- It states the simulated population frame, the run date, and a caveat.
- Its wording says it is simulated or directional (e.g. “in a simulated panel, 58% preferred…”).
- Citation text generated by the page appends “(simulated, directional)”.
We will not:
- Present a simulated figure as a population estimate, market size, or census fact.
- Publish simulated willingness-to-pay as a price point. Price-sensitivity results are directional screens only.
- Publish conversion-rate, sales, or “will buy” forecasts from simulations.
- Mix simulated and published figures in one statistic.
These rules are enforced by a publish gate in our CMS: a Market Insight cannot be published if any simulation statistic lacks its frame, run date, or caveat, and purchase-certainty wording must be reviewed and justified by a person before publishing.
Sample frames and run dates
The population frame describes who the simulated panel represents (for example, “urban adults aged 18–44 in Indonesia, Vietnam, and the Philippines”) and, where relevant, the number of synthetic respondents. The run date is when the simulation was executed. Published statistics show the date of the source publication.
Screening for mode collapse
Language-model-based simulations can collapse towards a single typical answer, understating real-world variety. Before a Market Insight is published, a reviewer checks simulated distributions for implausibly low variance or single-option dominance, and removes or re-runs affected results. This check is recorded with each publish.
Freshness and refreshes
Market Insights are refreshed roughly monthly. Each refresh re-checks sources for newer data, re-runs or adds simulations, updates the “Last updated” date, and adds a dated note to “What changed”. Statistic links stay stable across refreshes wherever the statistic is kept.
Limitations and known biases
- Simulations inherit gaps and biases in the public data used to calibrate them, including under-represented groups.
- Stated preferences in a simulation can differ from what people actually do, especially for price and new categories.
- Published sources differ in definitions and timing; we do not harmonise figures across sources unless stated.
- Each Market Insight lists its own limitations in its Methodology section.
Corrections and contact
If you find an error, an outdated source, or a statistic that is mislabelled, tell us through the contact form with the page link and statistic. We correct confirmed errors, note the correction in the page’s changelog, and update the Last updated date.
Want to run your own simulation? Start with a free quick concept test or preferences-validate.