Simile Raises $100M for Behavior-Prediction AI

A young AI company called Simile has raised $100 million to build systems that can forecast human decisions—what customers may buy, how audiences may react, and even what questions could surface on earnings calls. For leaders tracking where “agentic” AI is headed next, this is a clear signal that the market is moving beyond generation toward simulation, and it’s exactly the kind of shift we spotlight at the ai world summit 2025 / 2026.

The $100M round and what Simile is building

Simile’s new $100 million financing is designed to accelerate a model that predicts human behavior by running simulations built from AI agents that represent real people’s preferences. The round was led by Index Ventures, with participation from Bain Capital Ventures, A* and Hanabi Capital, and it also included well-known AI figures such as Fei-Fei Li and Andrej Karpathy; the company did not disclose its valuation.

The company says its ambition is straightforward to state and difficult to execute: anticipate what a person is likely to do in a given situation by creating a “sandbox” where many plausible outcomes can be tested before a business commits real money, time, or reputational risk. Instead of leaning only on focus groups or small-sample surveys, Simile’s approach aims to provide a faster way to explore customer intent, sentiment, and choice under different conditions.

From the lens of the ai world organisation, this matters because behavior prediction sits at the center of modern business strategy—marketing, product, pricing, merchandising, and investor communications all depend on how people interpret information and act on it. That’s why the ai world summit, ai world summit 2025, and ai world summit 2026 themes consistently circle back to applied AI decisioning, practical enterprise outcomes, and responsible deployment—core pillars across ai world organisation events and ai conferences by ai world.

How “predicting behavior” works in practice

Simile emerged from stealth after spending roughly seven months developing its model, and it says the system was trained using interviews with hundreds of people about their lives, alongside historic transaction data and text from scientific journals focused on behavioral experiments. The intent of mixing these inputs is to produce agents that don’t just sound plausible, but behave in ways that resemble real populations—at least well enough to support business decisions.

In practical terms, simulations are only useful if they can be queried like a business tool: “What happens if we change the offer?”, “Which message reduces churn?”, “How will a specific customer segment respond to a policy update?”, or “What concerns will analysts raise on the next call?” Simile explicitly points to forecasting likely earnings-call questions and predicting how corporate announcements may be received by analyzing prior calls and research, which signals an early focus on high-stakes enterprise workflows rather than consumer novelty demos.

This is also where evaluation becomes the real battleground: forecasting human behavior is notoriously hard, and accuracy depends on whether the simulated agents generalize beyond the data they were trained on. At the ai world organisation, we frame this as the “trust gap” problem: enterprises need clear benchmarks, repeatable tests, and guardrails before they will treat simulated outcomes as decision-grade evidence, and these are the kinds of implementation questions that come up repeatedly at the ai world summit 2025 and ai world summit 2026 sessions.

Enterprise use cases: retail shelves to earnings calls

Simile’s examples show why investors are excited: if an AI system can reliably predict what people will choose, it could reshape how companies plan inventory, target promotions, and design store experiences. As one concrete test case, CVS Health has been testing Simile’s service to inform what items to stock and how to display products in stores, according to the company.

That retail scenario is only the start, because “behavior prediction” can be applied anywhere decisions are made under uncertainty: personal finance journeys, subscription retention, onboarding flows, and customer support escalation paths. The same underlying capability—simulating how different personas respond to different stimuli—can also help leaders anticipate second-order effects, such as whether a change that boosts short-term conversion might increase long-term churn or complaints.

Simile also highlights corporate communications, including preparing for analyst questions on earnings calls and forecasting reactions to announcements by analyzing historical context. For many executives, that’s compelling because it reframes communications as a measurable system: companies can test messages, identify likely points of pushback, and rehearse responses before speaking to markets.

For the ai world organisation, the big takeaway is that “agentic AI” is becoming operational: it’s being positioned as a planning instrument, not just a chat interface. That’s why our editorial direction around the ai world summit and related ai world organisation events keeps emphasizing real enterprise adoption playbooks—what data is required, how models are validated, and how teams translate AI output into accountable decisions—topics that continue to drive attendance at ai conferences by ai world.

Risks and responsible deployment: what leaders must pressure-test

Behavior-prediction systems raise a different risk profile than standard generative AI, because the output can influence decisions that affect pricing, credit, hiring, healthcare access, or public opinion. Even if a simulation is statistically “useful,” it may encode bias from its inputs, oversimplify human context, or fail when conditions change—especially when models are trained on narrow samples, incomplete transaction histories, or literature that does not reflect current populations.

Data governance is another pressure point: Simile describes training on interviews and transaction data, which implies that consent, anonymization, and usage rights are not optional extras but foundational requirements. Enterprises evaluating tools like this should insist on clear documentation of data sourcing, retention practices, and whether outputs can be traced back to sensitive attributes or re-identification risks.

Then there’s the organizational risk: a simulation can look persuasive even when it’s wrong, so leaders need processes that prevent “automation bias” (people deferring to model output because it sounds confident). In ai world organisation programming at the ai world summit 2025 and ai world summit 2026, we push a simple governance standard for high-impact AI: simulations should inform decisions, not replace accountability, and teams must maintain a measurable feedback loop from real-world results back into model evaluation.

What this means for 2026 AI strategy (and for AI World events)

The Simile round shows that venture capital is rewarding application-layer AI that promises measurable business leverage—especially tools that compress research cycles, reduce experimentation costs, and improve decision quality. It also suggests a near-term competitive shift: companies that can run credible “what-if” experiments on customers, markets, and communications may out-iterate peers that rely on slower testing methods.

If you’re building an AI roadmap, treat this as a prompt to expand your definition of “AI capability” from content generation to decision support: segmentation, propensity modeling, scenario planning, and communications intelligence are now converging into one agentic toolkit. The practical question for leaders is not “Will simulation be perfect?” but “What level of accuracy is good enough for which decision, and what controls prevent misuse?”

This is where the ai world organisation comes in: our upcoming global summits are designed to connect builders, enterprise leaders, and investors around exactly these applied questions, with a focus on practical strategies you can implement immediately and a network spanning leaders from many countries. If you’re tracking behavior-prediction AI as a category, plan to bring your use case, your constraints (data, compliance, timelines), and your evaluation checklist to the ai world summit, ai world summit 2025, and ai world summit 2026 conversations—because the winners in 2026 won’t be the teams with the flashiest demos, but the teams who can deploy responsibly at scale via ai world organisation events and ai conferences by ai world.