Simulant turns real data into living simulations — so decisions can be tested before they become real.
Research has always looked backwards: what people said, what they did, what happened after the move was made. Simulant makes research forward-facing. We turn real respondent data into living simulations of markets, audiences and communities, so organisations can test possible futures before committing to one. But where others generate AI personas, Simulant is building the infrastructure for synthetic research: respondent models grounded in real human evidence, probabilistic answers that preserve uncertainty, validation against held-out human data, and clear signals for when a question can — and cannot — be trusted. It is not a faster survey or a synthetic focus group. It is a new decision layer between imagination and reality, built to make the most important moves testable before they become irreversible.
Simulant turns real respondent data into living population models, so organisations can test how people may respond before decisions enter the world.
Hear the language beneath the numbers. Explore grounded synthetic verbatims that reveal concerns, motivations and trade-offs across different audiences.
Measure the shape of opinion before it hardens. Test support, preference, intent, trust and demand across the groups that matter.
Convene the room the decision needs. Bring together any audience, community or stakeholder segment and observe how views shift in conversation.
Model behaviour, not just opinion. Use observable patterns — votes, purchases, signals and repeated choices — to forecast what people may do next.
Stanford researchers built generative agents from interviews with 1,052 real people — the agents matched their subjects’ survey answers 85% as accurately as the people matched their own answers two weeks later.
Park et al., “Generative Agent Simulations of 1,000 People” (Stanford University, 2024).
Read the paperModel how your strategy plays across your audience before you run it.
Find the message that lands — and the reason it does.
Pre-test ads, hooks, and hero assets before the spend.
Model reach, resonance, and the blind spots ahead.
Stress-test a statement against the room before you send it.
See where you win, where you lose, where you're ignored.
Rehearse the market's reaction before you ship.
Watch how you're seen move over time and segment.
Follow how a story spreads, mutates, and takes hold.
Simulant is not AI roleplay. It is a probabilistic inference layer built on respondent-level survey data: every model is anchored to a real respondent, every estimate carries measured uncertainty, and the system is backtested against held-out human answers.
Read the full white paperEach respondent model corresponds 1:1 to a real respondent in a source dataset. The observed record — answers, codebook labels, and survey weights — remains the source of truth.
For each new question, the engine assembles the observed responses most relevant to it and records the exact fields supplied to the model — an auditable evidence trail for every inference.
The engine estimates a probability distribution over the valid response options rather than forcing a single answer, and repeats inference independently to measure run-to-run stability.
Responses are inferred for every eligible respondent, then aggregated using the source study's survey weights — preserving the population structure of the original sample.
Real answers are hidden from the model, predicted, and compared with ground truth. Accuracy and calibration are scored at both respondent and population level.
Before a study runs, each question is scored for how strongly the source evidence supports inference. Weakly supported questions are flagged for human fieldwork.
Simulate your decisions against a representative model of the people you need to reach — and see not just whether it lands, but why.