How Microsoft built an AI coach to scale personalized sales training

What would happen if you built an AI version of your CEO to coach thousands of sellers?

At Microsoft, that question emerged from a much larger challenge: How to rapidly upskill our customer and partner-facing teams in a way that is scalable, role-specific, and grounded in real business conversations. To bring this to life, we focused on how our sellers could practice and apply these skills in real-world scenarios.

One example of this approach is Agent J.ai. This agent is an on-demand, AI-powered coach we designed to enhance the skills and performance of 64,000 of our Microsoft Commercial employees serving in sales roles. And it’s not just any AI-powered coach—it’s modeled off of Judson Altoff, CEO of our commercial organization.

For our sales, partner, and service teams, this project represented a shift in how we approach skilling internally. We’re not just delivering training on Microsoft AI solutions; we’re using AI itself to power how those skills are learned in practice.

At Microsoft, this is what Customer Zero looks like in real-world scenarios—using our own AI solutions to solve a core business challenge: Helping customer-facing teams build skills for real customer and partner conversations at scale.

We’re building and testing AI tools inside the organization, observing what consistently works, and sharing practical insights with our customers.

Upskilling employees on Microsoft AI solutions

Upskilling thousands of employees at scale—and giving them customized instruction that fits their specific needs—is a tremendous challenge for any organization. Teams have limited time, and there’s the issue of making sure the content is up to date and in line with organizational strategy. These teams also need to access accurate, specific information across a vast content repository.

Our internal AI Transformation team had already embarked on a global campaign to educate and train our customer and partner-facing teams on Microsoft AI solutions. They had developed job aids, demos, and videos tailored to sales scenarios like account planning, opportunity qualification, and customer meeting prep. They’d also launched campaigns, Learning Days, and contests like “The Road to 60,” which encouraged teams to reach 60% daily active usage of our Al solutions.

These initiatives showed positive results, with a 68% boost in customer planning efficiency for an account executive, and a 62% gain in agile workflows for a business program manager. But the teams also realized that AI itself could help scale and customize this training even further. That’s where Agent J.ai came in.

Upskilling with a familiar AI coach

According to Stacey Herod, a senior learning manager here at Microsoft, project members gathered focus groups of employees in varying roles across the commercial organization to help understand the issues that customer and partner-facing teams face. They pinpointed challenges with speaking the language of the customer, establishing executive presence, connecting solutions to business outcomes, and understanding how to best serve as a trusted advisor to customers.

The team wanted to develop a coaching experience that could simulate real customer conversations and reflect how strong sellers actually operate in the field.

At the center of that experience is a familiar voice modeled after Althoff. For customer and partner-facing teams, Althoff’s approach represents how experienced leaders define the vision and share deep expertise that speaks to our customers’ needs.

“It started as a fun experiment—replicating Judson as a mini version of himself, where you could always have him by your side, coaching you in real time,” Herod says. “But it quickly revealed an opportunity to scale executive-level guidance. Now, customer and partner-facing teams can bring their own scenarios and engage with a trusted voice—making learning more personal and impactful.”

Instead of a one-to-many model, we can now interact with and learn from that perspective on demand.

Learning isn’t about just accessing information; it’s about understanding how experienced leaders navigate conversations, connect solutions to business outcomes, and build trust with customers and partners. Modeling the experience on that perspective helped translate those behaviors into a scalable, practical format.

Althoff was fully onboard with the project and closely involved with its creation, adhering to the company’s principles around responsible and ethical AI use. We also knew that having executive buy-in and sponsorship is critical when it comes to increasing adoption of tools like Agent J.ai across the company.

“We want people to feel comfortable practicing conversations with artificial intelligence, so they can refine their approach and be prepared ahead of a customer meeting, not practicing with their customers.”

Stacey Herod, senior learning manager, Microsoft

Rethinking how skills are built with AI

At Microsoft, this shift is already underway. Agent J.ai is part of a broader ongoing effort to rethink how skills are built with AI, and our employees are trusting the experience. Importantly, conversations that our employees are having aren’t tied to individual identities, rather the content is curated to the specific needs around upskilling employees. That sense of privacy made it easier for sellers to engage openly and use it without hesitation.

“There’s behavioral change happening at the agentic level—people’s level of comfort using agents,” Herod says. “We want people to feel comfortable practicing conversations with artificial intelligence, so they can refine their approach and be prepared ahead of a customer meeting, not practicing with their customers.”

Practice is central to this model.

“Agent J.ai fundamentally changes skilling—from static training to dynamic, scenario-driven coaching. It adapts in real time to each seller’s context, enabling hundreds of roles and limitless customer scenarios to be addressed instantly—something our industry has never achieved at this scale.”

Jennifer Wheeler, senior learning manager, Microsoft

Role-playing helps simulate realistic, highly customized conversations that customer and partner-facing teams might have with a certain type of client or executive. For example, an account executive might need to prepare for an upcoming conversation with a chief security officer.

Our AI coach’s training must be able to recognize that context.

“Agent J.ai fundamentally changes skilling—from static training to dynamic, scenario-driven coaching,” says Jennifer Wheeler, a senior learning manager at Microsoft. “It adapts in real time to each seller’s context, enabling hundreds of roles and limitless customer scenarios to be addressed instantly—something our industry has never achieved at this scale.”

An AI agent trained on curated content, Microsoft’s customer engagement and coaching frameworks, and a trusted persona could finally deliver what teams lacked—authentic conversation practice tailored to real scenarios, industries, and executive audiences.

Launching and refining our AI coach

We rolled Agent J.ai out quickly, with a focus on iterating based on real-world use. From the start, members of the cross-functional team prioritized capturing feedback and continuously refining the experience based on how sellers use the tool in practice.

“We are seeing the rise of voice-first subject matter expertise agents across industries, and Agent J.ai has demonstrated what’s possible in the Frontier.”

Max Felker, principal product manager, Microsoft

The team began the Agent J.ai project in earnest in March 2025 and shipped the first version just five months later in July. Input from super users helped test specific scenarios, while broader anonymous forms surfaced key insights.

One example: In the user experience, we discovered that using overly realistic or animated avatars for the AI coach could be perceived by users as gimmicky and might prove more expensive to produce. A simple photo avatar ended up doing the trick.  

As a voice-based coaching agent, Agent J.ai is designed for spoken, two-way conversations with users.

“We are seeing the rise of voice-first subject matter expertise agents across industries, and Agent J.ai has demonstrated what’s possible in the Frontier,” says Max Felker, principal product manager for Microsoft.

Incorporating updated industry trends and live events

Staying current is critical for customer-facing teams, especially when conversations shift quickly based on new announcements and industry trends.

“I had a couple of people who said, ‘This is incredible. This saved me hours, or weeks, and I was able to have it in 5 minutes.’ That’s something we never thought would be possible.”

Stacey Herod, senior learning manager, Microsoft

Typically, pulling information together for real customer conversations took time. Powered by Microsoft Azure and Azure AI, Agent J.ai can incorporate announcements from events like Microsoft Ignite, so the AI coach could help prepare sales teams for questions about the latest news.

“I had a couple people who said, ‘This is incredible. This saved me hours, or weeks, and I was able to have it in 5 minutes,’” Herod says. “That’s something we never thought would be possible.”

Others reported saving up to 50% of their standard prep time, not to mention more than $80 million in deals that were influenced through refined strategy and approach.

In the end, the team was laser-focused on making something valuable and accessible for their sales teams.

“If the Agent wasn’t useful, then our learners were never going to come back and use it,” says Herod, a seller in the past herself. “You have one shot to impress sellers. We’re a tough community.”