In this episode of Sourcery, host Molly interviews Joon Sung Park, co-founder of Simily, an AI company valued at $2 billion after raising $300 million total from investors including Greenoaks and Index Ventures.

Joon details the evolution of generative human simulation, starting from the landmark 2023 Stanford 'Smallville' experiment to Simily's present-day platform powered by 40 million panel participants. They explore the mechanics of predicting human behavior, enterprise applications across retail and finance, strategies for eliminating survey bias, and the grand vision of simulating complex societal challenges like climate change coordination and Universal Basic Income.

From Stanford's Smallville to 85% Behavior Prediction

The 2022-2023 'Smallville' experiment placed 25 generative AI agents with memory, planning, and reflection architectures into a virtual town, observing emergent, unprompted social behaviors.

By late 2024, Simily validated its architecture on 1,000 demographically representative Americans, achieving over 85% accuracy in predicting real human behavioral choices.

"In late 2024, for the first time in the field, we proved we could predict people's behavior with over 85% accuracy using these agents." - Joon

The Science of Human Behavior and Choice Architecture

A primary challenge in behavioral modeling is bridging the gap between stated intentions in surveys and actual actions taken in real-world contexts.

Survey responses typically follow a normal distribution, whereas actual real-world action requires overcoming activation energy, behaving more like an exponential curve.

"What actually governs people's behavior is choice architecture—the available options and their opportunity costs at the exact moment of decision." - Joon

Simily's Enterprise SaaS Platform and Global Panels

Simily operates as a SaaS platform enabling enterprise clients like CVS and Banco Itaú to conduct rapid concept testing, conjoint analysis, and messaging optimization.

The company maintains data partnerships covering over 40 million real people globally, allowing custom demographic panel simulations to be provisioned within 24 hours.

Simily explicitly excludes minors and individuals unable to give informed consent from its data collection processes.

To ensure reliable decision-making, Simily outputs a confidence calibration score where metrics above 85-90% serve as actionable evidence for high-stakes decisions.

Simulating Macro Societal Dynamics and Future Scenarios

Behavioral simulation allows organizations to model second-order systemic impacts, such as how EV marketing strategies unintentionally alter consumer perception of non-EV vehicles.

Joon highlights the potential of simulation to address generational coordination problems like climate change policies and Universal Basic Income without multi-year physical trials.

Simily prioritizes identifying causal mechanisms over simple correlations, giving decision-makers effective levers to actively shape future outcomes.

Building Simily's Team and Field Deployment Strategy

Simily's 100-person team includes ~20-25% research and methodology scientists from Stanford labs alongside engineering and field deployment talent.

Prominent angel investors and advisors include AI leaders Fei-Fei Li and Andrej Karpathy, stemming from shared academic roots at Stanford.

"Field deployment strategy creates stickier, more resilient enterprise relationships, which will define the simulation market as it matures over the next two years." - Joon

Resources Mentioned

Journal Notebook for Data People | Data Scientists, Data Analysts, Data Engineers, Students, Teachers

es. Building Simily's Team and Field Deployment Strategy Simily's 100-person team includes ~20-25% research and methodology scientists from Stanford labs alongside engineering and field deployment ta