Large Behavioral Models: The Next Class of Artificial Intelligence

​The data that drives business decisions is not linguistic. It is transactional, purchases, clicks, cancellations, supply chain decisions made under constraint. These are observed choices, made by real actors, in real time. Large language models were not designed to learn from this kind of sequential, non-verbal signal.

​A new model class addresses that gap.

​Large Behavioral Models are trained on sequences of human action. Each transaction functions as a token. Each decision carries signal. Rather than summarizing historical patterns, the model learns to predict what happens next.

​Language captures what people say. Behavior captures what they do. The gap between those two data types represents a meaningful and largely untapped source of enterprise value.

​This pattern has precedent. Harvey identified an opportunity in legal reasoning before the broader market. Cursor did the same with code generation. Perplexity, with search.

​That dynamic appears to be forming again around behavioral data.

​A workshop at Harvard Business School with Erik Guander and Rickard Brüel Gabrielsson from Unbox AI on how Large Behavioral Models are used by Fortune 100 companies today.