Aug 22, 2026 · Latent Space: The AI Engineer Podcast · 5 min read

Simile: Behavior Foundation Models for Decision Simulation

Simile leverages behavior foundation models to simulate human decision-making with 85% accuracy, moving beyond market research to causal strategy. The company uses randomized controlled trials and deep behavioral data to help enterprises predict outcomes and optimize complex, multi-variable business environments.

Technology Business Strategy Entrepreneurship

The takeaways

Behavioral Data Beats Attitudinal Surveys. Simile prioritizes causal behavioral data from randomized controlled trials over self-reported attitudes to model how people actually act, not just what they say.
85% Accuracy in Human Prediction. By training on deep individual data, Simile achieves 85% accuracy in replicating human behavior, significantly outperforming general-purpose LLMs which often drop to 20-30% in niche contexts.
Simulation Enables Causal Strategy. Unlike prediction tools that forecast outcomes, simulation maps the step-by-step path to a goal, allowing leaders to identify counterintuitive interventions that shape future results.
Two-Model Architecture for Scale. The platform uses distinct population-level and individual-level models to balance broad statistical trends with granular, persona-specific behavioral nuances for precise targeting.
TAM Exceeds Market Research. Simile positions its technology as a tool for all human decision-making, not just market research, suggesting a total addressable market far larger than the $100 billion research industry.
Real Stakes Validate Behavior. Data collection relies on experiments with real-world consequences, such as actual product deliveries, to ensure the behavioral signals captured are genuine rather than hypothetical.

The Shift from Prediction to Causal Simulation

Simile is redefining the AI landscape by moving beyond generative content to behavior foundation models that simulate human decision-making. While traditional market research relies on attitudinal data—what people say they will do—Simile focuses on behavioral data—what people actually do. By leveraging randomized controlled trials (RCTs) and deep qualitative interviews, the company builds digital twins that replicate individual human behavior with 85% accuracy. This approach addresses a critical gap in current AI: general-purpose large language models (LLMs) often fail to capture the nuanced, non-rational, and context-specific social physics of human interaction, particularly in niche markets where their accuracy can drop to 20-30%.

Strategic Value: Shaping the Future

The core value proposition of Simile is not merely predicting outcomes but enabling causal strategy. Decision-makers in enterprise, CPG, and policy sectors often know the direction of a trend but lack the insight into the specific interventions required to alter that trajectory. Simulation allows users to test counterfactuals and identify non-intuitive steps that lead to desired goals. For example, a marketing campaign for an EV might inadvertently harm overall brand perception; simulation reveals these second-order effects before deployment. This shifts the utility of AI from a reactive tool to a proactive strategic partner, capable of modeling complex, multi-agent environments where individual actions aggregate into societal or market-level outcomes.

Data Architecture and Scaling

Simile’s technical edge lies in its data architecture, which categorizes inputs into three buckets: qualitative interview data for texture, observational behavioral data for statistics, and causal mechanism data from RCTs for predictive power. The company employs a dual-model system, training both population-level and individual-level models to ensure scalability without losing granularity. As compute costs decrease and data collection scales to tens of millions of users, Simile is observing early scaling laws in simulation performance. The long-term vision extends beyond commercial applications to solving "wicked problems" like climate change and democratic stability, positioning simulation as a critical infrastructure for societal decision-making. By grounding AI in real-world behavioral stakes, Simile is building a new category of technology that informs every decision made about humans, for humans.

Key insights

General-purpose LLMs are insufficient for modeling niche human behavior because they lack deep, individual-specific behavioral data, often resulting in low accuracy for specific market segments.

Technical Limitation →

Impact: Enterprises relying on generic AI for market insights risk making flawed decisions in specialized verticals where behavioral nuance is critical.
Causal mechanism data, derived from randomized controlled trials with real stakes, is significantly more valuable for prediction than attitudinal survey data or passive observational logs.

Data Strategy →

Impact: Companies should invest in active experimental data collection over passive scraping to build more robust predictive models for consumer behavior.
Simulation provides a step-by-step path to a goal rather than a single predicted outcome, allowing for the discovery of counterintuitive strategic interventions.

Strategic Utility →

Impact: Leaders can use simulation to test complex, multi-variable strategies that would be too costly or risky to implement in the real world.
The total addressable market for behavior simulation is not limited to the $100 billion market research industry but encompasses all human-centric decision-making processes.

Market Opportunity →

Impact: Investors and founders should view simulation as a foundational layer for enterprise software, with potential applications in policy, healthcare, and urban planning.
Simile’s dual-model architecture, separating population-level trends from individual-level traits, allows for scalable yet granular insights that balance statistical power with personalization.

Product Architecture →

Impact: This approach enables businesses to target specific micro-segments with high precision while maintaining a broad understanding of market dynamics.

Action items

Audit current market research methodologies to identify reliance on attitudinal data and begin piloting behavioral experiments with real-world stakes.

Impact: This shift will improve the accuracy of consumer insights and reduce the gap between stated intent and actual behavior.
Evaluate the use of simulation tools for high-stakes strategic decisions, such as product launches or policy changes, to test counterfactual scenarios.

Impact: This can prevent costly mistakes by revealing second-order effects and non-intuitive paths to success before full-scale deployment.
Invest in collecting causal mechanism data through controlled experiments rather than relying solely on passive data scraping or surveys.

Impact: Causal data provides a more robust foundation for predictive models, especially in complex or niche markets where general trends are less reliable.
Develop a dual-model approach for customer analytics, separating broad population trends from individual-level behavioral traits.

Impact: This allows for more precise targeting and personalization while maintaining the statistical validity of broader market insights.
Explore partnerships with research institutions or platforms that provide high-quality, pre-registered randomized controlled trial data to enhance AI model training.

Impact: Access to rigorous experimental data can significantly improve the accuracy and reliability of behavior foundation models.

Quotes

“Most decision makers, what they want to know is how can we shape the future?”

“The models that we are trying to create are models that are as dumb as I am.”

“Simulation is a tool for human decision making.”