CVS Health has turned to artificial intelligence to better understand consumer behavior, using AI-generated “agentic twins”—digital replicas of real people—to simulate human decision-making. Last fall, the company analyzed data from 400,000 AI agents to explore challenges such as why many patients fail to fill or complete their prescriptions. This effort was part of a broader move to leverage AI simulations for market research and consumer insights.
The technology behind this approach comes from Simile, a startup based in Palo Alto, California, founded last year by Joon Sung Park while he was completing his Ph.D. in computer science at Stanford University. Simile’s platform creates AI agents that mimic human behavior by drawing on a large and continuously expanding database of simulated individuals. These agents allow companies to test reactions to new products and assess brand perception without the time and expense associated with traditional market studies.
“If you’re able to simulate the world, you can basically test out countless interventions,” said Park, who is also Simile’s chief executive and co-founder. The startup’s client roster now includes CVS Health, Deloitte, and Wealthfront, among others. On Thursday, Simile announced it had secured $200 million in new funding, pushing its valuation to $2 billion. This latest round was led by Greenoaks and included investments from prior backers such as Index Ventures, Hanabi, Bain Capital Ventures, CVS Health Ventures, and Definition. This funding follows a $100 million raise less than six months earlier.
Simile’s development is rooted in research modeling human behavior in interactive environments inspired by simulation games like The Sims. Park collaborated with co-founders Michael Bernstein and Percy Liang, both Stanford computer science professors renowned for foundational contributions to computer vision and AI research, respectively. Lainie Yallen, formerly of AI startups Hebbia and Valence, also co-founded the company.
The company initially created its core behavioral model through in-depth interviews with 1,000 participants representing a broad demographic cross-section. These insights were used to develop AI agents whose responses have since been validated against new data, reaching an accuracy rate ranging from 85 to 99 percent depending on the context and demographics, according to Simile’s head of product, Mihika Kapoor. Further partnerships with organizations like Gallup have helped scale the data to include input from several million people worldwide.
Simile’s platform allows clients to simulate and analyze specific population groups through a web interface, generating models that can include anywhere from hundreds of thousands to millions of AI agents. Pricing is linked to the size of these simulated populations. As the field of AI-driven market research expands, competitors such as Helm, Artificial Societies, Rehearsals, and Aarui have also emerged.
Looking ahead, Park envisions moving beyond static survey-style models toward “dynamic simulations” that track AI agent interactions over time, enabling long-term product testing and behavioral analysis. This evolving technology aims to provide companies with deeper, faster insights into consumer responses in a cost-effective manner.