Boston Consulting Group is teaching its customer-facing AI, known as Jamie, not only to copy top-performing sales behaviors but also to avoid the missteps that frustrate customers. That dual approach — training on both high-performing calls and poor interactions — aims to boost sales effectiveness while tightening coaching and quality control for clients.
Japjit Ghai, a managing director at BCG X, described the work on a recent company podcast and in a follow-up interview. BCG uses recordings and transcripts supplied by clients, plus its own research and a firm’s internal knowledge, to build models of effective seller behavior and to flag tactics that tend to fail.
Rather than cloning a single salesperson, the technology looks for recurring patterns across many conversations to understand which techniques consistently produce positive results, Ghai said. The system then applies those lessons in two ways: it helps an AI agent steer live interactions and it produces feedback for human sellers.
Saving money: US personal savings rate drops to 2.6% in April, lowest in years
Montgomery County awards $2.75M in grants for 16 community projects
- Data sources: client call recordings and transcripts, internal BCG research, and company-specific knowledge.
- Learning goals: identify repeatable, successful behaviors and recognize practices that alienate buyers.
- Outputs: AI-guided responses during customer conversations and individualized performance scorecards after calls.
After a call, a salesperson can receive a tailored report showing strengths and weaknesses based on how they navigated the conversation — for example, whether they asked the right questions, addressed objections promptly, or used language that resonated. Over time, every interaction feeds back into the model, helping the agent refine its guidance and the coaching it provides.
The approach is already reflected beyond consulting. Vercel, a cloud platform for web developers, built an agent modeled on its most successful sales representative and subsequently reduced its SDR team to a single human overseer, according to company comments last year. David Totten, Vercel’s vice president of global field engineering, told reporters that mimicking top performers is a longstanding practice; modern AI simply accelerates the process.
That acceleration carries immediate implications. For businesses, AI agents trained this way promise more consistent customer experiences and scalable coaching. For employees, the technology can shift roles toward supervision and exception handling rather than routine outreach. For customers, the hope is fewer clumsy or irritating interactions — provided the models are well tuned.
There are also operational and ethical considerations. Training on customer-call data raises questions about consent, data governance, and security. Companies sharing recordings with vendors must ensure they have the appropriate permissions and safeguards in place, and buyers should expect clear disclosures about how their information is used.
Two trends to watch:
- Firms will increasingly use internal call archives as a strategic asset for building AI coaching tools and virtual sellers.
- Organizations will need tighter policies around data sharing and auditing to prevent models from learning harmful or biased behaviors.
BCG frames Jamie as a continuously improving capability that learns from aggregated behaviors rather than mimicking any single salesperson. That distinction matters: by emphasizing patterns instead of personalities, the consultants say the system can scale desirable skills while reducing the risk of repeating poor tactics.
Correction, May 14, 2026 — This article was updated to clarify that BCG trained the agent using customer-service data provided by its clients.
