A startup wants companies to stop asking real people what they think, and ask fake ones instead. Simile has raised $200m at a $2bn valuation to build “agentic twins,” AI stand-ins for real consumers that firms can survey by the million.
Greenoaks led the round, the New York Times reported, with Index Ventures and others returning. It is a fast turn. The year-old startup raised $100m less than six months ago.
Surveying people who do not exist #
Simile’s pitch is that it can predict how people will behave by simulating them. Companies use it to test a new product or gauge how a brand lands, faster and cheaper than a traditional market-research study.
The customers are real even if the respondents are not. Last autumn, CVS Health used 400,000 agentic twins to work out how to get patients to take their medicine. Deloitte and the finance app Wealthfront are clients too.
“If you’re able to simulate the world, you can basically test out countless interventions,” said Joon Sung Park, Simile’s 32-year-old chief executive. He called it “a window into humanity.”
Real humans built the fake ones #
The founders come straight from the research that started this field. Park built Simile from a 2023 Stanford paper on AI agents that mimic human life, a kind of AI version of The Sims. His co-founders include Percy Liang, who helped coin the term “foundation model.”
The model itself started with real people. Simile ran two-hour interviews with 1,000 representative participants, then teamed with Gallup to add input from a few million more. It claims its answers are 85% to 99% accurate, depending on the group.
Pricing follows the crowd. Simile charges partly by how many agents it models, from hundreds of thousands to millions. “The larger the panel, the higher the price,” Park said.
A bold, unproven idea #
The premise is contested. Those accuracy figures are Simile’s own, and whether AI stand-ins truly capture how humans behave is exactly what critics doubt. Rivals such as Helm, Artificial Societies and Aaru are chasing the same idea, part of the wider hunt for measurable value from AI.
There is irony in it, too. Simile leans on real human data to manufacture the fake kind. It does so on a web already flooding with AI agents, where proving you are a real person is its own business. Its stated goal, per Pulse 2.0, is to model all eight billion people on earth. It is the sort of promise enterprise buyers will judge on results.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.