
AI prediction of consumer liking from sensory properties for product development and optimization
Description
Food and beverage companies rely on trained sensory panels and consumer tests to predict product success. These tests take one to three months, cost $10,000 to $100,000, and need expert knowledge of statistics. This project uses a Human Intelligence + AI + Data approach to predict consumer liking in about one week, for about $2,500, with an accuracy of 75% or better. Building on a 2025 seed grant and a coffee study accepted in npj Science of Food, we will (1) expand a FAIR Sensory and Consumer Database across food, beverage and dish categories, and add nutrition and composition data where useful; (2) find consumer preference segments, select “reference tasters” who represent each segment, and validate them against traditional tests; and (3) train and continuously retrain machine learning models that predict consumer liking from sensory data and link it to business variables such as sales.
Team

Jean-Xavier Guinard
Principal Investigator

Ilias Tagkopoulos
Co Principal Investigator