CVS Health wanted to know how to get people to take their medicine.
Millions of patients never fill their prescriptions, and many don’t finish them. It was one of many questions that the company sought to answer and improve for its customers. So last year, it turned to data collected from 400,000 people to look for answers.
There was just one catch: They weren’t actually people. Instead, they were “agentic twins” of real people created with artificial intelligence (AI).
The company that handled the data was Simile, a 1-year-old startup in Palo Alto, California. Simile says its technology can predict human behaviour by tapping into its growing database of AI agents. That allows clients to test the reaction to new products or evaluate how their brand is being perceived by consumers. And, Simile promises, they can do so faster and cheaper than by running traditional market research studies.
“If you’re able to simulate the world, you can basically test out countless interventions,” said Joon Sung Park, Simile’s CEO and co-founder. “This is sort of the window into humanity.”
It’s a bold sales pitch, and it appears to be working. In addition to CVS, Simile’s fast-growing client list includes the consulting group Deloitte and Wealthfront, a financial services app.
On July 30, Simile is announcing US$200mil (RM818mil) in new funding at a US$2bil (RM8.18bil) valuation. Less than six months ago, the company raised US$100mil (RM409mil). Greenoaks led the latest round, which also included existing investors Index Ventures, Hanabi, A*, Bain Capital Ventures, CVS Health Ventures and Definition.
Park, who is 32, launched Simile last year while finishing his doctorate in computer science at Stanford University. The blueprint was a research paper he co-wrote in 2023 about AI agents simulating human behaviour, such as dating or going to work, in an interactive environment modelled after the popular game Sims.
“The first sentence that Joon shared with me was to simulate the entire planet,” said Shardul Shah, an Index Ventures partner. “As a sci-fi nerd – like, I love ‘Hitchhiker’s Guide to the Galaxy’ – this just lit my imagination on fire.”
Park’s co-founders include Michael Bernstein and Percy Liang, both Stanford computer science professors. Bernstein is one of the authors of the seminal ImageNet paper, which laid the foundations for modern computer vision technology, and Liang was the lead researcher for a paper that coined the term “foundation model” in 2021. Lainie Yallen, who previously held go-to-market roles at the AI startups Hebbia and Valence, is also a founder.
Simile built its initial model by conducting two-hour interviews with 1,000 participants, who were selected to represent the broader population. The company used that data to design its core model for human behaviour, then tested its responses against new answers from the human panellists.
Simile has since teamed up with Gallup and other partners to supplement its data with input from a few million people around the world. Simile’s responses are now between 85% and 99% accurate, depending on the scenario and the population surveyed, according to the company’s head of product, Mihika Kapoor. The startup says it already has enough data to tap into the AI-simulated opinions of very specific population groups.
The company’s product is a web interface in which users can select different populations that they want to run a simulation on. Simile generates revenue based in part on the number of agents that it models for its customers. That can range from hundreds of thousands to millions of agents.
“The larger the panel, the higher the price,” Park said.
Simile is not alone in the new field of AI simulation research. Startup competitors include Helm, Artificial Societies, Rehearsals and Aaru.
Park envisions expanding beyond survey-based market research. He said he planned to develop “dynamic simulations,” in which AI agents are measured over time, such as testing products over the course of a week. – ©2026 The New York Times Company
This article originally appeared in The New York Times.
