TL;DR
Boston Consulting Group is training an AI sales agent called Jamie on both its best and worst sellers’ behaviours, using call transcripts and engagement patterns to build a system that coaches human salespeople and improves over time. Vercel has already taken the same approach further, cutting its 10-person SDR team to one human overseeing an AI agent. BCG earned $3.6 billion from AI consulting in 2025 (25% of $14.4 billion revenue) and is part of OpenAI’s Frontier Alliances for enterprise AI deployment.
Boston Consulting Group is doing something slightly unusual with its new AI sales agent. It is teaching it how to fail.
The agent, called Jamie, is being trained not only on the call transcripts, engagement patterns, and conversational habits of BCG’s strongest-performing sellers, but also on the behaviours that did not work, the approaches that fell flat, the moments where a human salesperson lost the room. In an industry that has spent the past two years talking almost exclusively about what AI can do, BCG appears to be equally interested in what it should avoid.
Japjit Ghai, a managing director and partner at BCG X, the firm’s technology build-and-design division, explained the approach on a recent episode of BCG’s podcast, The So What from BCG. “We trained the agent by studying the best sellers, their call transcripts, how they engage with customers, and teaching Jamie to do the same,” Ghai said. “We also trained Jamie not to replicate the worst seller experiences.”
How Jamie learns
In an interview with Business Insider, Ghai described the training process in more detail. Jamie draws on several sources: BCG’s own internal expertise, a client’s knowledge of its business, and the company’s existing sales calls and transcripts. Those recordings, Ghai said, represent a “repository of often underleveraged assets” that BCG can mine to identify “what good looks like.” The goal is not to clone any individual seller but to extract historical patterns that reveal which behaviours and engagement styles tend to resonate with customers, and which do not.
The system is also designed to coach the humans it works alongside. After a call, a seller can receive a personalised scorecard from Jamie showing what they did well and where they fell short, based on how they navigated the conversation. Each interaction generates more data that feeds back into the model, making Jamie what Ghai described as an “always on muscle” that improves continuously.
The framing is careful. BCG is not positioning Jamie as a replacement for its salespeople but as a distillation of collective institutional knowledge, an attempt to encode what the firm’s best people do instinctively and make it available to everyone. Whether that framing survives contact with the economics of AI deployment is another question. When you build a system that can replicate your best seller’s behaviour at scale, the pressure to need fewer sellers becomes structural, regardless of the original intent.
The Vercel precedent
That dynamic has already played out elsewhere. Vercel, the cloud platform for developers, took the same principle, training an AI agent on its top-performing sales development representative, and followed it to its logical conclusion. The company reduced its 10-person SDR team to a single human overseeing the AI agent. The remaining nine were moved to outbound prospecting roles rather than let go, but the message was clear: once an AI agent absorbs what your best person does, you need fewer people doing the same thing.
David Totten, Vercel’s vice president of global field engineering, told Business Insider in October that “modelling after the top-performing employees has always been a standard business practice. The difference now is that technology lets us accelerate it.” The company reported that the AI agent handles inbound qualification, filters spam, and routes leads, work that previously occupied an entire team.
The consulting bet
For BCG, Jamie sits within a much larger strategic wager on AI. The firm reported $14.4 billion in revenue for 2025, with 25% of that, roughly $3.6 billion, coming directly from AI-related consulting work, according to a disclosure first reported by Bloomberg in April. AI and technology services now account for more than 40% of BCG’s total revenue. The firm has expanded its workforce to 33,500 employees, adding AI engineers, data scientists, and IT architects alongside its traditional consulting teams.
BCG X, where Ghai is based, was created in late 2022 by merging three existing subsidiaries, BCG Digital Ventures, BCG Platinion, and BCG Gamma, into a single unit of more than 3,000 technologists, scientists, and designers. The division builds AI-driven solutions for clients including L’Oreal, BMW, and New York Life. Jamie, then, is not just a product BCG is building for its own use. It is also a demonstration of what it sells: the proposition that AI agents can learn from an organisation’s best performers and operationalise that knowledge at scale.
The broader consulting industry is moving in the same direction. OpenAI announced partnerships with BCG, McKinsey, Accenture, and Capgemini in February to help sell and implement its Frontier AI agent platform. Under these “Frontier Alliances,” each consulting firm is investing in dedicated practice groups and building teams certified on OpenAI’s technology. The consultants’ role is to help clients redesign workflows, integrate AI agents with existing systems, and manage the organisational disruption that follows.
What makes BCG’s approach to Jamie worth watching is the acknowledgement, still rare in corporate AI discourse, that training on success alone is insufficient. Every sales organisation knows that its best performers succeed partly through instinct, timing, and interpersonal chemistry, qualities that resist easy codification. Teaching an AI what not to do may prove more tractable than teaching it what to do, because failures tend to be more systematic than successes. The seller who talks over the client, who pushes a product before understanding the need, who mistakes persistence for persuasion, these patterns repeat. An AI that avoids them consistently might not match the best human seller on a good day. But it might outperform the average one on every day, which, for a firm that bills by the engagement, could matter more.
