Simile raises $100M to build AI digital twins of real humans

Before shipping a major redesign or rolling out a price hike, most companies rely on a familiar playbook: survey existing users, assemble focus groups, and cross their fingers. It is slow, costly, and finding the right people to talk to is often nearly impossible.

Simile thinks it has a faster way out. The AI digital twin startup just announced a $100 million funding round led by Index Ventures, with backing from Bain Capital Ventures alongside AI heavyweights Fei-Fei Li and Andrej Karpathy.

Instead of asking busy professionals to fill out forms, Simile creates personalized AI clones of real individuals to predict how they would react to new features, interfaces, or corporate decisions.

From the Smallville sandbox to enterprise software

If the premise sounds familiar, that is because Simile was founded by the computer scientists behind Smallville: Joon Sung Park, Michael Bernstein, and Percy Liang. Smallville was the viral Stanford experiment that dropped 25 autonomous AI agents into an RPG-style town to see if they could coordinate, socialize, and throw parties on their own.

Now they are applying that behavioral modeling to the enterprise. Built over seven months, Simile's core model was trained on detailed interviews with hundreds of people, augmented with transaction logs and scientific journal texts.

Index Ventures partner Shardul Shah noted that the startup partners directly with individuals to map out how they think and make choices. The goal is not generic roleplay, but answering what specific buyers will do and why.

Predicting analyst calls and testing UI updates

Companies are already putting the technology to work. Telstra, Australia's largest mobile operator, and healthcare giant CVS Health are among Simile's early roster of enterprise clients.

One common use case is interface testing. Before rolling out controversial UI changes, product teams can run the layout past simulated users to spot confusion or backlash early.

Simile is also pitching executive prep. CEO Joon Sung Park told Bloomberg that the model correctly anticipated eight out of ten analyst questions in a simulated earnings call test run. For executives facing unpredictable investor scrutiny, that kind of prep is pure gold.

Why it matters

We think Simile is tapping into one of the biggest bottlenecks in modern product development: access to elite feedback. Getting thirty minutes with a Fortune 500 CTO is brutally hard. If an AI twin can capture even seventy percent of that perspective, teams can iterate far faster.

That said, behavioral simulation carries real risk. Human decision-making is chaotic, emotional, and heavily influenced by context that rarely shows up in transaction logs or interview transcripts. If a product bet goes wrong because an AI focus group gave it a green light, who owns the fallout?

Still, the broader signal is clear. AI is moving beyond generating content and toward anticipating human reactions. If synthetic audiences prove reliable, corporate strategy will increasingly be tested on thousands of digital humans before a single real customer ever sees it.