Pomegra
Simile, a Stanford spinout, closed a $100M Series A led by Index Ventures to build AI systems that replicate human decisions at enterprise scale, emerging from seven months of stealth.
Key Takeaways:
- Index Ventures led the $100M Series A, joined by Bain Capital Ventures, Hanabi, A*, and angels Fei-Fei Li and Andrej Karpathy.
- Simile's platform creates AI digital twins trained on real interviews and transaction data to simulate consumer decisions.
- CVS Health is among early clients, using Simile for inventory planning and earnings call preparation before the company went public.
Lead
Simile raised $100 million in a Series A led by Index Ventures, the company announced in February 2026. The Stanford-based startup had operated in stealth for roughly seven months before the announcement. Joining Index in the round were Bain Capital Ventures, Hanabi Capital, and A\*. On the angel side, AI researchers Andrej Karpathy and Fei-Fei Li participated alongside Adam D'Angelo, Guillermo Rauch, and Scott Belsky. The company declined to disclose a post-money valuation.
What Does Simile Actually Build?
Simile creates AI digital twins - synthetic representations of real people built from interviews, transaction data, and behavioral research - that companies can query to forecast how customers or employees will respond to a given decision. The aim is to replace, or at minimum displace, traditional focus groups and market surveys with simulations that can be run millions of times in hours rather than weeks.
The company was founded by Joon Sung Park, whose Stanford PhD dissertation produced "Smallville," a project in which generative AI agents lived simulated lives, formed relationships, and coordinated behaviors with minimal human oversight. That research attracted significant academic attention and built the conceptual foundation for Simile's commercial direction. Park co-founded the company with fellow Stanford researchers Michael Bernstein, Percy Liang, and Lainie Yallen, each bringing backgrounds in human-computer interaction and large language model research.
Why Did Index Lead at $100M Into a Stealth Company?
Index framed the investment as backing "foundational infrastructure for decision-making in an AI-native world." That phrasing matters. It positions Simile not as a market research vendor but as something closer to an operating layer for enterprise behavior prediction - a larger and more defensible market claim.
The decision to commit at this size also likely rested on early enterprise traction. CVS Health had been testing Simile's platform for inventory decisions and earnings call preparation while the company was still in stealth. Fortune 100 customers before a public launch are an unusual signal at Series A stage. They shorten the distance between technical promise and commercial proof.
The stature of the angel roster reinforces the case. Karpathy and Li rank among the most credible technical voices in AI, and their presence carries practical weight with enterprise procurement teams evaluating a new and unfamiliar category. D'Angelo and Rauch add operational credibility. The list was selected to say something, not just supply capital.
What Is the Competitive Risk?
The market for synthetic consumer data is not empty. Survey platforms, qualitative research firms, and a growing number of AI-powered research tools compete for the same enterprise line items. The distinction Simile draws is fidelity: rather than asking respondents how they expect to behave, the system builds a model of who they actually are. That is a different claim, and it invites a higher standard of verification.
Foundation model providers are not standing still. As general-purpose models improve at behavioral simulation, the question of whether a specialized layer on top commands a durable price premium will only sharpen. Simile's own framing that its agents "fail like humans" is a creative reframe of a technical constraint - whether customers find that reassuring or unsettling will depend on context.
How Accurately Can AI Predict Novel Human Behavior?
This is the category's core problem, and no one has solved it cleanly. Digital twins trained on historical interviews and transactions can likely reproduce past decisions under familiar conditions. How they perform on genuinely novel stimuli - a new product category, a macro shock, a viral reputational event - is harder to assess and easier to oversell. Simile's ability to retain enterprise clients through high-stakes decisions will eventually provide that test better than any controlled demo.
Outlook
Simile enters the market with $100 million, a technically credible founding team, and early validation from at least one major enterprise client. The round buys time and talent to prove whether AI-generated simulations of human behavior can be accurate and consistent enough to influence consequential decisions at scale. Without a disclosed valuation, it is not possible to compare this round against any prior pricing - that figure may surface in the next raise. The category Simile is building remains genuinely unsettled, which is both the opportunity and the risk.
