Vectorial raises $3 million to simulate human behaviour

On 20 October Vectorial announced a $3 million seed round from Smilegate, SenseAI Ventures, Berkeley SkyDeck and Dent Capital. Teams use it to test product journeys, campaigns, brand messages and customer-facing AI agents against simulated users built by its SAPIENS behaviour model before going live.

Vectorial says SAPIENS is 30% more accurate at predicting behaviour than general-purpose LLMs and that simulations reproduced prior user-study findings with more than 85% accuracy. It reports modelling over 2.5 million real people and running over 30,000 simulations. Customers include GoodRx, MIDI Health, Everkind, boAt and Concrete Software. These figures are self-reported.

The money goes to new behavioural data partnerships and SAPIENS 2.0. A synthetic user is only as good as the real behaviour data behind it.

The relatively small seed round shows the field is still early, and much depends on whether behavioural data partners can supply diverse, high-quality signals rather than narrow samples.

User research is slow and expensive; synthetic users promise hours instead of weeks. The open question is whether they capture people's unpredictable, emotional, context-dependent choices or just repeat the average. Behavioural economics shows decisions are full of systematic biases, and systems that model them could be a real advantage.

As companies deploy customer-facing agents, testing needs change. Seeing how a chat agent handles an impatient, undecided or angry customer takes thousands of scenarios that cannot each be run with real people. Synthetic users offer a scalable way to test agents against many human types before launch, which is where Vectorial positions itself.

• Use simulation as a first filter, then decide on real user data.
• Collect your own behaviour data on where customers drop off.
• Name the behaviour: loss aversion or choice overload, not just "cart abandonment".