CEO and co-founder Amit Gal-Or told TechCrunch that human behavior is inherently uncertain, but underlying tendencies can help estimate the likelihood of particular actions.
“Behavior is not deterministic; it’s constantly probabilistic,” Gal-Or said.
The system is designed to produce probabilities for different decisions, rather than claim to know exactly what someone will do. According to 42AI, those estimates can remain useful when a person encounters unfamiliar circumstances because certain behavioral characteristics may persist across situations.
To build these predictions, the company analyzes interactions that reveal how individuals respond to different experiences.
In financial services, that information can include account registration, identity verification, customer support conversations, money transfers, and trading activity. The system examines factors such as communication patterns, sentiment, transaction timing, and market conditions to identify behavioral tendencies.
42AI converts those observations into mathematical representations that are processed by a transformer-based world model. The model then simulates possible decisions under different conditions and estimates their likelihood.
A separate language model generates explanations for the resulting predictions.
Co-founder and CTO Ehud Ezra said the system is trained to maintain consistency in its representation of individual behavior. For example, a person identified as generally cautious about risk should not suddenly be modeled as highly risk-seeking without supporting evidence.
The approach draws on research spanning psychology, neuroscience, and genetics. During development, 42AI consulted specialists in psychology and post-traumatic stress disorder, along with Dan Shechtman, the recipient of the 2011 Nobel Prize in Chemistry.
The company initially studied thousands of individuals using psychological assessments and behavioral exercises. It subsequently incorporated government and biobank datasets representing more than 2 million people.
Gal-Or said the customer datasets available to the company now cover nearly 30 million individuals.
In commercial deployments, 42AI processes existing customer information, including transaction histories and support interactions. Gal-Or maintains that the system does not require personally identifiable information to generate predictions.
“The model never knows who the person is,” he said.
However, the technology still builds individual behavioral profiles by connecting information from a person's interactions, raising questions about how those profiles are used and protected.
42AI reports strong results in its financial applications, although the company's performance figures require some distinction.
Gal-Or claims the model reaches 98% accuracy in forecasting trading activity. He clarified that this figure measures whether trading flows remain within predicted ranges, rather than whether the system correctly anticipates every individual transaction.
For predictions involving trading direction and volume, the company reports F1 scores between 0.8 and 0.85. The metric evaluates a model's ability to balance accurate classifications against false positives and missed predictions.
The founders also acknowledge limitations. The system cannot account for every experience influencing a person's decisions, and predictions become less reliable when important circumstances are missing.
For instance, a trader might change their investment behavior after receiving advice from someone else, introducing information that the model cannot observe.
Despite those constraints, 42AI says it is already generating commercial revenue.
Gal-Or reported that the startup works with 10 major brokerages, which use the technology to support liquidity management, margin calculations, and hedging strategies. He said revenue exceeds seven figures but declined to provide a specific amount.
Customers can pay a fixed fee or compensate 42AI based on a share of the financial value attributed to its platform.
Gal-Or claimed that one brokerage generated approximately $30 million in additional value through improved dynamic hedging.
The company's $62.5 million seed financing was provided over three years by its founders and Infinity Group, meaning the total does not represent funding raised entirely from outside investors.
Infinity Group was founded by Amir Gal-Or, the father of 42AI's CEO, who also serves as the startup's chairman and a co-founder. The company intends to pursue additional financing.
Although financial services represent its initial commercial focus, 42AI is evaluating applications in gaming, cybersecurity, and fraud prevention.
Its longer-term ambitions include helping AI agents and robots interpret human behavior. However, Ezra acknowledged that expanding into new industries requires additional training to account for their particular rules and operating conditions.
The technology also introduces questions about potential conflicts of interest.
A financial institution capable of estimating how individual customers might trade could potentially use that information for its own advantage. Ezra acknowledged the concern, while the founders cited regulatory requirements, monitoring procedures, and an internal ethics committee as safeguards.
They did not specify technical restrictions that would prevent customers from using behavioral predictions in ways that disadvantage the individuals being analyzed.
42AI's immediate commercial results suggest there is demand for more sophisticated behavioral forecasting in financial services. But its broader claims remain dependent on how reliably the model performs outside the environments where it has been trained and deployed.
The company has reported measurable prediction performance and financial benefits for brokerage customers, while acknowledging that incomplete information can undermine its forecasts. Its expansion into other industries will provide further opportunities to assess whether the same approach can predict human decisions across substantially different situations.
This analysis is based on reporting from TechCrunch & the tech buzz.
Image courtesy of 42AI.
This article was generated with AI assistance and reviewed for accuracy and quality.
