How Does Nubank Test New Features So Fast?

David Vélez says Nubank has treated itself since the beginning as a technology company that happens to be in financial services, so data and analytics drive every decision, and Nubank's own foundation model cut the cost of putting a feature into a model far below the old process of spending six months to add a new feature and several weeks to test one in a model.

Traditional banks competed on banking branches. Consumers picked a bank with a branch nearby, so effort went into real estate, physical infrastructure, brands, and trust, which financial services still has to protect. Nubank did not spend energy on an offline world and treated technology and data as the comparative advantage. Before AI, adding a new feature took six months, and testing one feature in a model took several weeks of work. With the foundation model, the team can test many approaches much faster. An experiment as small as an email with purple letters or yellow letters now costs far less. Every decision in the organization can be automatized and made by a model versus a human.

AI lets Nubank do the same data-driven work at a much larger scale. The culture was already in place because technology and data were the way to differentiate from the start.