42AI raises $62.5 million to predict how people behave

From interactions to probabilities

42AI’s model is not meant to replay a person’s past choices. CEO and co-founder Amit Gal-Or told TechCrunch the company is trying to measure persistent psychological tendencies that influence decisions even as circumstances change. Co-founder and CTO Ehud Ezra said the system estimates the probability of a particular choice, rather than treating behavior as predetermined.

For investors and traders, the signals can include KYC procedures, service registrations, support calls, money transfers and stock trading. The company says it first analyzes speech tone, voice and other interactions with customer support or sales teams, including how often someone calls. It then adds information about when and how that person buys and sells assets, alongside market conditions.

The observations are converted into mathematical representations of personal traits and passed to a transformer-based world model. A separate language model generates explanations for its predictions. In simulations, the system is trained to preserve a person’s tendencies: if someone usually avoids risk but the model starts classifying them as risk-seeking, Ezra said, the training objective has failed.

The company says it developed the approach over three years, combining psychology, neuroscience and genetics with advice from specialists in psychology and post-traumatic stress disorder, as well as 2011 Nobel Prize in Chemistry laureate Dan Shechtman.

Its data collection began with psychological assessments and behavioral scenarios involving thousands of people. 42AI later added government and biobank data covering more than 2 million people. Gal-Or said customer datasets have since grown to nearly 30 million people.

A strong number with a narrow meaning

The headline performance figure is 98% accuracy for predicting market trades. But Gal-Or said that number reflects whether a stream of trades falls within a predicted range, not whether the system correctly forecasts each individual trade.

For trade direction and volume, he said, the model consistently scores between 0.8 and 0.85 on F1, a measure of classification performance that accounts for false positives and missed events. The system also has an obvious limit: it cannot know everything happening in a person’s life. Ezra said accuracy falls when the model lacks important context, such as a change in strategy prompted by a friend or colleague.

42AI says those 10 large brokerage firms use its system to manage liquidity, margin requirements and hedging risk. The startup is onboarding more clients and already earns more than seven figures in revenue, though Gal-Or did not give an exact amount. Customers pay either a fixed fee or a percentage of the benefit generated by the platform. Gal-Or said one brokerage gained about $30 million from dynamic hedging.

The profile is the hard part

The company says its models do not need personal data and do not know who a person is. Yet its method depends on linking a customer’s actions across settings to build an individual behavioral profile. That distinction may matter technically, but it does not answer the practical question of what a brokerage can infer about a client.

I think the more consequential uncertainty is not whether the model can predict a trade within a range. It is who gets to act on the prediction, and whose interests that action serves. Ezra acknowledged concerns that a brokerage could use insight into customers’ behavior for its own advantage. The founders pointed to regulation, monitoring and an internal ethics committee, but did not explain what would stop clients from using the technology to gain an edge over the people it analyzes.

The funding story is also less straightforward than the headline suggests. The $62.5 million in seed funding came over three years from the founders and Infinity Group; Gal-Or said it was not all raised from external investors. Infinity was founded by his father, Amir Gal-Or, who is also 42AI’s chairman and a co-founder. The startup now plans to raise additional funding.

42AI is also exploring gaming, fraud prevention and cybersecurity, and wants eventually to help robots and AI agents understand people. Ezra said each new field would require training the model on its own rules and context. That caveat points to the broader challenge: a system built to infer people’s motives is only as useful—and as risky—as the setting in which someone chooses to use it.