Simile secures $200 million to evaluate AI "twins" of actual consumers.

      A startup is urging companies to cease asking real individuals for their opinions and instead consult artificial ones. Simile has secured $200 million at a $2 billion valuation to develop “agentic twins,” AI representations of real consumers that businesses can survey in large numbers.

      Greenoaks led this funding round, as reported by the New York Times, with Index Ventures and other investors participating again. This marks a quick turnaround for the one-year-old startup, which raised $100 million less than half a year ago.

      Simulating non-existent respondents

      Simile’s proposition is that it can forecast consumer behavior by simulating individuals. Companies utilize this technology to test new products or assess brand reception more quickly and affordably than traditional market-research methods.

      While the respondents may not be real, the customers are. Last fall, CVS Health employed 400,000 agentic twins to determine how to encourage patients to adhere to their medication regimens. Other clients include Deloitte and the finance app Wealthfront.

      “If you can simulate the world, you can essentially test numerous interventions,” stated Joon Sung Park, Simile’s 32-year-old CEO, referring to it as “a window into humanity.”

      Real humans created the artificial ones

      The founders are directly from the research that initiated this field. Park established Simile based on a 2023 Stanford paper regarding AI agents that replicate human life, a form of AI akin to The Sims. His co-founders include Percy Liang, who contributed to the term “foundation model.”

      The model itself originated from real people. Simile conducted two-hour interviews with 1,000 representative participants and collaborated with Gallup to gather insights from millions more. It claims that its results are 85% to 99% accurate, depending on the demographic.

      Pricing is based on the volume of simulated agents. Simile charges partly according to the number of agents modeled, ranging from hundreds of thousands to millions. “The larger the panel, the higher the price,” Park noted.

      A daring, untested concept

      The underlying premise is debated. The accuracy figures are provided by Simile itself, and there are concerns regarding whether AI substitutes genuinely reflect human behavior, as critics question. Competitors like Helm, Artificial Societies, and Aaru are pursuing similar concepts as part of the broader quest for measurable AI value.

      There is also irony in the approach, as Simile relies on actual human data to create artificial counterparts. It operates in a realm already saturated with AI agents, where establishing oneself as a real person is an industry on its own. According to Pulse 2.0, its ambition is to model all eight billion people on the planet. This is the kind of promise that enterprise clients will evaluate based on outcomes.