AI Consumer Research Simulator debuts at Google Cloud hackathon

The project builds on ideas from Dentsu's People Model, People Research and AI For Growth Talk, and draws on Stanford University's AI for Public Benefit Lab and research published in Nature in 2026. Its statistical foundation combines variables from Japanese public sources including the Census, the Employment Status Survey and the Family Income and Expenditure Survey, and the project discloses the citation sources and manual assumptions behind its six consumer-value segments in the UI and README.

The team designed the system with governance safeguards, setting the participant-consent flag to false by default and rejecting import requests without consent confirmation at the API level. In the public judging environment, uploading real personal data or deleting the base demo data returns an error, and any fallback to a mock model is shown clearly in the UI to prevent mistaking it for real AI output.

The app is built around seven AI agents, including one that converts real qualitative interview records into seven psychological and behavioral dimensions with verbatim quotes, and another that randomizes the order in which control and treatment conditions are shown to counter anchoring bias. A validation agent removes pseudo-random error entirely and splits human data in half to evaluate prediction accuracy without post-hoc calibration. For concept testing, users input a new-product idea, price and description, and the system outputs purchase intent, price acceptance, usage scenarios and improvement suggestions.