Simile vs. Aaru: Which Human Behavior Simulation Platform Is Right for You?

Behavior simulation went from research curiosity to one of the most heavily funded corners of AI in under two years. Simile and Aaru are the two names that come up most often, and they are frequently mentioned in the same breath. They are not the same product. This guide compares them on what they model, the data behind them, what buyers use them for and what to check before you trust any simulation. Full disclosure: we build Cairo Rosetta, a behavior model for organizations, so we include it where it is relevant and say plainly where it is not.

Simile vs. Aaru at a glance

Simile: a foundation model of human behavior

Simile grew out of the 2023 Stanford Generative Agents research, in which 25 AI characters in a simulated town formed relationships and organized a party on their own. Its co-founders are Joon Sung Park, Michael Bernstein and Percy Liang. A follow-up study by Park and colleagues in 2024 built agents from two-hour interviews with 1,052 people and found that they replicated those people's survey answers about 85% as accurately as the people replicated their own answers two weeks later.

The company emerged from stealth in February 2026 with a $100 million Series A led by Index Ventures, with Bain Capital Ventures, A*, Hanabi Capital, Fei-Fei Li and Andrej Karpathy participating. Five months later it raised a $200 million Series B at a $2 billion valuation.

What makes Simile distinctive is what it optimizes for. Frontier language models are trained to be helpful, rational agents, which makes them poor stand-ins for real people. Simile post-trains on interviews, transaction data and registered behavioral experiments specifically to recover human biases, inconsistency and causal patterns. In other words, it wants its model to make the same mistakes people make.

Best for: large enterprises that need to test product, messaging or strategic decisions against simulated customers or stakeholders with individual-level fidelity.

Aaru: simulating entire populations

Aaru was founded in March 2024 by Cameron Fink, Ned Koh and John Kessler. Its software, Lumen, generates large numbers of AI agents from public and proprietary data and simulates how specific demographics or geographies will respond to future events and decisions. Accenture Song committed to integrating Lumen into its AI offering, and Aaru lists partners such as EY and Interpublic Group.

In December 2025, Aaru raised a Series A led by Redpoint Ventures at a reported $1 billion headline valuation, with a multi-tier structure that put the blended valuation lower. Its use cases sit at the population level: pricing, political and election forecasting, disaster-response planning and adoption-rate prediction, often replacing studies that would take weeks with results in days.

Best for: organizations that need fast, population-scale answers to questions traditionally handled by surveys, panels and polls.

The real difference: individuals vs. populations

Both companies predict human behavior, but they answer different questions:

• Simile asks: how would this kind of person, or this specific segment, react, and why? Its value is fidelity and causal texture.
• Aaru asks: how will a population distribute its reactions? Its value is speed and scale across demographics and geographies.

If you are testing a pricing change across a national market, population simulation is the natural fit. If you are deciding between two product concepts for a high-value customer segment and want to understand the reasoning, individual-level simulation is.

Where neither fits: behavior inside organizations

Both platforms are built mostly for consumers, voters and markets. Predicting how your own employees will react to a reorganization, a new policy or a leadership message is a different problem. The data that matters is not purchase history or polling. It is how each person tends to decide, collaborate and handle pressure, and how those tendencies interact in a team.

That is the problem Cairo Rosetta is built for. It takes a structured psychometric profile of a person (in Cairo, from 80+ behavioral indicators) plus a description of the situation, and predicts the most likely choice. In our first public experiment, Rosetta reached 60% accuracy on 30 held-out real decisions, versus 56.7% for the best frontier LLM, with about 41% lower latency. With only 30 cases, that accuracy gap is not statistically significant, and we report it that way. Read the full results in Cairo Rosetta vs. frontier LLMs, and the category overview in what is a large behavior model.

How to evaluate any behavior simulation vendor

Frequently asked questions

What is the difference between Simile and Aaru?

Simile builds a foundation model of human behavior focused on high-fidelity simulations of individuals and segments. Aaru runs multi-agent simulations of whole populations to forecast how demographics and markets will respond.

Is Simile or Aaru more accurate?

There is no independent head-to-head benchmark. Simile's founding research reported agents that matched people's survey answers about 85% as well as people match themselves. Ask each vendor for accuracy on held-out data in your own use case.

Can Simile or Aaru predict employee behavior?

They can simulate people in general, but neither is designed around employee profiles or organizational change. Organizational behavior models like Cairo Rosetta use psychometric profiles of real employees for that purpose.

Do synthetic simulations replace real research?

Not yet. They are best used to narrow options quickly and decide what to test with real people. See what is synthetic research.

Sources

1. SiliconANGLE, AI digital twin startup Simile raises $100M in funding, February 12, 2026.
2. Dealroom, Simile company profile.
3. Latent Space, Why simulation is taking over, August 2026.
4. Park, J. S., et al., Generative Agent Simulations of 1,000 People, 2024.
5. NeuronFeed, Aaru company profile.
6. Simplify, Aaru (Lumen and Accenture Song).
7. Cairo Labs, Rosetta Arena, Experiment 1, October 2026.