20VC: The Best AI Companies Have Unique Data Acquisition Strategies | Will Simile Kill Kalshi, Polymarkets and NASDAQ | How to Sign Fortune 500 Companies As Customers in Weeks with Joon Sung Park, Simile

What It Covers

Joon Sung Park, founder of Simile, explains how his Stanford research on AI agents simulating human behavior in a virtual town evolved into a $300M-funded company that helps Fortune 500 enterprises like CVS predict and shape future consumer behavior through validated human simulation models with 85% behavioral accuracy.

Key Questions Answered

• •Defensible Data Strategy: AI companies in this generation must build data acquisition strategies competitors cannot replicate. Simile collects behavioral data — transactions, observational records, and randomized controlled trials — rather than relying solely on web-scraped text. The critical differentiator is running AB tests showing how behavior changes under different conditions, creating causal training data rather than purely correlational observational data that only supports prediction, not decision-shaping.
• •Causal Modeling Over Prediction: Enterprises do not ultimately want predictions — they want to change outcomes. Simile's models are built around counterfactuals: showing clients what happens if they take action A versus B. CVS does not need to know Frappuccino sales will drop; they need to know which intervention prevents it. Building causal reasoning into simulation models, not just forecasting, is what drives enterprise willingness to pay.
• •Enterprise Sales Speed: Fortune 500 deals can close in under three months when the pain is acute enough. Simile demonstrated this by replicating, in two minutes, findings from consulting studies that took three to six months to produce. Founders targeting enterprise should lead with live demonstrations that replicate a prospect's existing research — this collapses skepticism and compresses sales cycles dramatically compared to traditional software procurement timelines.
• •Synthetic Panels Replacing Human Panels: Simile's models predict individual behavior with 85% accuracy compared to how people replicate their own responses. Synthetic panels will exceed the current human panel market within three years because they remove budget, time, and scale constraints. Currently, roughly 95% of hypotheses organizations want to test never get tested due to cost and logistics — simulation removes that ceiling entirely for consumer research.
• •Simulation Compute Economics: Single simulation sessions will cost $10–20M to run within two to three years, yet enterprises and governments will pay $100M for the output because the decisions informed are worth multiples of that. This mirrors the trajectory of inference scaling in frontier models. Founders building in simulation should price against decision value prevented or generated, not against research budget line items, to capture appropriate value.
• •Researcher-to-Founder Transition Signal: When evaluating academics starting companies, the key signal is whether they are driven by impact or attached to a specific problem. Researchers married to a problem often build science projects; those driven by real-world impact find commercially viable problems. Additionally, look for team members who are the consistent common denominator across multiple career stages — this signals extreme ownership and the ability to reinvent under new conditions.