Simile's Joon Sung Park on using AI to model human behavior at 85% accuracy

At ICONIQ Frontier, Joon Sung Park, CEO of Simile, shared how the company is building models that predict how people will behave in a given environment.

The approach grew out of Joon’s research at Stanford. He studied roughly 1,000 Americans, using hours of interview and behavioral data to build a digital twin of each participant. Two weeks later, the real participants and their twins completed the same surveys and behavioral studies. The twins matched their responses at more than 85%.

Simile now combines interview, survey, and behavioral data to model both individuals and populations, with a separate confidence model indicating how much to trust each result. That can help turn work like focus groups, simulated earnings calls, or large-scale conjoint studies into queries.

For one Fortune 10 retailer, that meant moving from roughly 100 studies a year to more than 2,000 a quarter.