Explosive growth of AI agents requires a shared context base

A platform to unify customer context

Birdie, a Silicon Valley startup founded by Brazilians, aims to transform the various signals generated by customers into a shared context, enabling both humans and AI agents to make decisions based on a common understanding. The company consolidates information from conversations, research, complaints, usage data, customer segmentation, and other consumer interactions into a single platform.

“If ten agents need to understand the same customer before performing ten different tasks, how much of that understanding really needs to be rebuilt ten times?” asks Alexandre Hadade, CEO and co-founder of the company. “The problem isn’t just the cost of redoing it. It’s the risk of ten different answers for the same customer.”

The tool’s value becomes even more critical when AI agents start participating in decision-making. If a customer cancels a purchase or service, product, customer experience, and support teams may come up with different explanations for the same case, each based on a system that doesn’t communicate with the others.

Birdie calls this a lack of context governance. The startup’s proposed solution is to create a database of information that can be used by different systems, with a record of the data’s origin and how it was used.

‘The more autonomy we give to AI, the more important the quality of the context behind it becomes’ – Alexandre Hadade, CEO and co-founder of Birdie

From analysis to decision

This information is already being used by companies that handle large volumes of customer interactions.

At KOHO, a Canadian financial services company, Birdie’s platform began analyzing a larger portion of consumer interactions. According to the startup, this helped reduce the structure dedicated to evaluating customer service and training employees by 80%. The estimated savings are around $350,000 per year.

At Brazilian fintech Neon, analysis of interactions identified problems that were evolving into regulatory complaints. Based on these signals, the company changed procedures while still in customer service. In two quarters, according to Birdie, the rate of regulatory complaints fell by 44%.

These examples demonstrate how information generated in customer service can be used outside of it. A complaint may indicate a product problem. A conversation may point to a process failure. A change in satisfaction may signal an issue that has not yet appeared in traditional indicators.

As the number of agents in companies increases, this information begins to feed a larger number of systems. The discussion, therefore, is not just about how intelligent each agent is, but about the quality of the context they will have to make decisions.

“We don’t believe companies should stop building their agents. On the contrary: they will build more and more,” says Hadade. “But our vision is that when they create the next agent, it doesn’t have to start from scratch to understand the customer.”