SIMILI, an AI startup from Stanford, uses advanced generative AI and voice-based interviews to create highly accurate simulations of human behavior for commercial applications like retail, finance, and polling. Their technology enables companies to predict customer actions and analyst questions with impressive accuracy, rapidly gaining traction with major clients despite being a young company.
The video discusses SIMILI, an AI startup spun out of Stanford University, which is focused on predicting human behavior using advanced generative AI models. The company’s approach centers on creating agents and simulations that can mimic real human actions and preferences in a wide range of contexts. Unlike traditional models that have struggled with accuracy and generalization, SIMILI leverages both structured and unstructured data, including in-depth interview data where people share their life stories and preferences, all collected with consent.
A key innovation from SIMILI is the use of AI interviewers that conduct one-on-one, voice-to-voice conversations with individuals. This allows the company to gather rich, nuanced data that was previously difficult to utilize. The collected information is then used to train AI agents that can represent real people in simulations, providing insights into how humans might behave in various scenarios. This methodology aims to create more accurate and generalizable behavioral models.
SIMILI’s technology is already being applied in commercial settings, particularly with Fortune 10 companies across sectors such as retail, personal finance, consumer packaged goods, and polling. For example, CVS has used SIMILI to create hundreds of thousands of digital agents representing their customers, enabling simulated focus groups to better understand customer pain points and optimize store layouts. Polling companies like Gallup are also collaborating with SIMILI to build digital panels of simulated human respondents.
Another notable application is in the simulation of corporate earnings calls. SIMILI’s models have demonstrated the ability to predict with 80% accuracy the questions that analysts are likely to ask CEOs and CFOs during these calls. This predictive capability suggests that AI could soon anticipate and prepare for key business interactions, potentially streamlining processes that currently rely on human intuition and experience.
Despite being a relatively new company—officially founded a year ago and with only seven months of model training—SIMILI has rapidly secured commercial deals and demonstrated impressive results. The company attributes its swift progress to a combination of leading academic researchers from Stanford and experienced product and engineering professionals from companies like Figma and Hebbia. This blend of cutting-edge research and practical product development has enabled SIMILI to quickly bring advanced AI behavioral modeling to market.