BCG trains its Jamie AI agent on mistakes, not just successes

Boston Consulting Group has built its own AI sales agent and named it Jamie. But it trains the agent in a decidedly unconventional way: not only on the successes of the company's strongest salespeople, but also on their mistakes and failures.

How Jamie learns

The classic approach to training AI in sales is to analyze best practices. Jamie does that too: it analyzes call recordings, patterns of customer interaction and the conversational habits of BCG's best salespeople. But the key detail that sets this project apart is that the agent also studies behavior that did not work.

Instead of simply copying only winning strategies, Jamie gets a full picture of which approaches led to refusals and lost deals. That includes attempts that looked reasonable on paper but failed in a real conversation with a customer.

Why learning from mistakes may be smarter

Analyzing successes alone is often not enough, because experienced salespeople often understand intuitively why certain techniques do not work. They simply stop using them, without realizing why or explaining their reasons. With BCG's approach, Jamie sees this hidden information explicitly.

The agent can identify patterns in failures that people do not notice:

• Specific phrases that increase customer objections
• Objection timing at which the customer is less receptive to the solution
• Patterns where an aggressive tone kills a potential deal
• Mistakes in lead qualification early in the call
• A lack of empathy in response to specific objections

The philosophy of learning through failure

BCG's approach differs from typical machine learning, where a system learns mainly from positive examples. Here the company has taken a deeper, more philosophical view: failures often teach more than victories.

This echoes the old principle of learning through failure used in sports, martial arts and military training. Champions remember their defeats more vividly than their wins, because failure forces them to rethink strategy.

"Failure is the best teacher if you know how to dissect it," say

sports psychologists.

What this means for the future of AI in sales

AI sales agents are becoming less mechanical and more human-like. Instead of simply following a "best practices" template, machines are now learning to actively avoid the traps that await even experienced salespeople. That could make AI in sales more realistic, more predictable and closer to the experience of a seasoned human manager.

This approach may also save companies from repeating the same mistakes. Even the best salespeople make errors, but their knowledge usually stays in the head of one person or a small team. With Jamie, those lessons become available for training the entire business, scaling to thousands of employees.

This is especially relevant for BCG, where a call has to do more than sell a consultation: it has to build trust. Jamie will learn not only how to close a deal but also how not to lose it at critical moments in the conversation.

What this means

AI sales agents are entering a new phase of development, moving from blind copying of successes to actively studying and preventing mistakes. This may prove a key moment in machines finally learning the way people do: not only through ups and downs, but through systematic analysis of failure. When companies start teaching AI on their own failures, they build systems that grow wiser with every mistake.