Synthesia has been one of the more visible names in the shift from traditional e-learning toward AI-assisted training. For years, the company’s pitch has centered on turning training content into something employees can consume quickly: video-based lessons, scripted scenarios, and avatar-led explanations that reduce production overhead while keeping the experience consistent. But today’s launch signals a meaningful change in direction. Instead of focusing primarily on what learners watch, Synthesia is pushing harder on what learners do.
The company introduced AI Roleplay Sessions, an enterprise training offering designed around live practice of workplace conversations with AI avatars. The core idea is simple to describe and harder to execute well: employees should be able to rehearse realistic interactions—difficult feedback, customer escalations, onboarding conversations, performance discussions, conflict resolution—then receive feedback that is specific enough to be actionable. In other words, the platform aims to convert training from passive viewing into measurable coaching.
What makes this launch notable isn’t just that it uses AI avatars. Many tools already let users “talk” to an AI character. The differentiator here is the training structure: roleplay sessions are built to be repeatable, scored, and analyzed so companies can evaluate whether training is actually improving performance over time. That emphasis on measurement is where the product starts to look less like a novelty and more like an enterprise learning system.
A move from content to practice
Video learning has a ceiling. Even when videos are interactive in small ways—quizzes, branching questions, scenario prompts—they still tend to keep the learner in a spectator role. Synthesia’s earlier work leaned into that model: generate training videos at scale, keep messaging consistent, and reduce the cost of producing new content for every team or region.
AI Roleplay Sessions changes the center of gravity. The learner isn’t just absorbing guidance; they’re expected to perform. During a session, an employee interacts with an AI avatar in a simulated workplace conversation. The avatar responds in real time, creating a back-and-forth dynamic that more closely resembles the unpredictability of actual work than a pre-recorded lesson ever can.
This matters because workplace communication is rarely about memorizing a script. It’s about adapting tone, choosing words carefully, responding to objections, and maintaining clarity under pressure. A roleplay format forces those skills into the open. If the learner says something vague, the conversation stalls. If they respond too aggressively, the avatar escalates. If they fail to ask clarifying questions, the avatar may steer the interaction toward confusion or incomplete outcomes. The training value comes from the friction of real conversation rather than the comfort of a one-way explanation.
Live coaching, not just conversation
The product’s headline features include feedback, scoring, and analytics. But the more interesting question is what “feedback” means in practice. In many AI chat experiences, the output is generic: a few suggestions, a summary of what the user “should have said,” and perhaps a motivational note. That kind of feedback can be helpful, but it often doesn’t translate into improvement because it lacks specificity and doesn’t connect to a measurable rubric.
Synthesia’s approach is positioned as coaching that can be evaluated. During the roleplay, the system provides feedback and scoring tied to the conversation itself. That implies the platform is not only generating responses, but also assessing the learner’s performance against criteria relevant to the training objective. For example, a session focused on handling a customer complaint would likely evaluate whether the learner acknowledges the issue, asks appropriate questions, demonstrates empathy, proposes next steps, and avoids escalation. A session focused on giving feedback might assess whether the learner uses constructive language, stays focused on observable behavior, and checks for understanding.
Scoring is particularly important because it turns coaching into a loop. Without scoring, training can become subjective: managers may feel the learner “seemed better,” but there’s no consistent way to compare sessions across teams, locations, or time. With scoring, companies can track progress, identify patterns, and determine whether training interventions are working.
Analytics then extend that loop beyond the individual. If a company can see aggregated performance trends—where learners struggle most, which scenarios produce the lowest scores, how performance changes after training—then the platform becomes a tool for learning operations, not just a training delivery mechanism.
Repeatable and scalable training
One of the quiet challenges in enterprise training is scaling quality. A great roleplay program in one department can become inconsistent when rolled out across multiple teams. Human coaching helps, but it’s expensive and difficult to standardize. Video libraries help with consistency, but they don’t always create the same level of practice.
Synthesia is explicitly aiming for repeatability and scalability. AI Roleplay Sessions is designed so organizations can deploy training across teams without needing to recreate scenarios from scratch each time. That’s not just about saving time; it’s about ensuring that learners are practicing against comparable conditions. If two employees complete the same roleplay scenario, the evaluation should be consistent enough that the score reflects performance differences rather than differences in scenario design.
This is where AI avatars can be more than a visual layer. If the avatar’s responses are grounded in the scenario context and the training objective, then the conversation becomes a controlled environment. The learner’s choices drive the outcome, but the simulation remains within boundaries that make it suitable for assessment.
In enterprise settings, that balance—freedom for the learner, structure for the evaluator—is often the difference between “cool demo” and “deployable system.”
Why workplace conversations are a natural fit
Workplace communication training is one of the most obvious use cases for conversational AI, but it’s also one of the hardest to do well. The reason is that workplace conversations are multi-dimensional. They involve not only what is said, but how it’s said: tone, pacing, empathy, clarity, and the ability to handle interruptions or pushback.
Roleplay sessions can simulate these dynamics because the conversation unfolds over time. Unlike a static quiz, the learner can’t simply select the correct answer from a list. They must craft responses that fit the moment. That creates a more realistic training environment for skills like:
Handling objections without escalating
Asking clarifying questions to reduce misunderstandings
Maintaining professionalism during emotionally charged interactions
Delivering feedback in a way that preserves dignity and encourages improvement
Navigating compliance-sensitive topics with appropriate language
The AI avatar becomes a stand-in for a colleague, customer, or stakeholder. The learner practices with someone who can respond immediately and vary the interaction based on what the learner does. Over multiple attempts, the learner can refine their approach, and the company can observe whether those refinements translate into higher scores.
The “measurable practice” thesis
Synthesia’s framing—turning training content into measurable practice—captures a broader trend in enterprise learning. Organizations increasingly want to justify training spend with evidence. They want to know not only that employees completed a course, but that the course changed behavior and improved outcomes.
AI Roleplay Sessions appears designed to support that demand. Feedback and scoring provide immediate signals to the learner. Analytics provide longer-term signals to the organization. Together, they create a training system that can be audited internally: which skills were targeted, how learners performed, and whether performance improved after training.
This is also where the platform could become valuable for compliance and governance. Many regulated industries require documentation of training completion and effectiveness. While roleplay scoring isn’t the same as certification, it can provide a structured record of competency practice. If a company needs to demonstrate that employees practiced specific conversation types—such as de-escalation, policy explanations, or customer handling—roleplay sessions can offer a more robust artifact than a watched video alone.
A unique take: coaching as a loop, not a one-time event
There’s a temptation in AI training products to treat roleplay as a single experience: you do one session, get some feedback, and move on. But the most effective training programs tend to be iterative. Learners need repetition, spaced practice, and opportunities to apply feedback immediately.
The “repeatable and scalable” language suggests Synthesia is thinking in terms of loops. Employees can likely revisit similar scenarios, compare scores across attempts, and improve specific weaknesses. Companies can also adjust training content based on analytics—adding more practice where performance is low, or refining scenarios where learners consistently misunderstand the objective.
That loop is where the product could differentiate itself in the market. Many AI tools can generate conversation. Fewer can reliably support a structured improvement cycle with consistent scoring and analytics that enterprises can trust.
What companies will care about next
As with any enterprise AI rollout, the next questions will revolve around implementation details and trust.
First, organizations will want to understand how scoring works and how transparent it is. Is the scoring based on a rubric aligned to training goals? Can companies customize evaluation criteria? Are there controls to ensure that the feedback is consistent and not overly influenced by minor phrasing differences?
Second, companies will want to know how scenario libraries are managed. Are roleplay sessions pre-built for common workplace needs, or can organizations create custom scenarios? The ability to tailor scenarios to internal policies, brand voice, and regional norms is often essential for adoption.
Third, there will be questions about integration. Enterprise learning platforms live inside larger ecosystems: HR systems, LMS tools, analytics dashboards, and identity management. If AI Roleplay Sessions can integrate with existing workflows, it becomes easier to deploy at scale and easier to measure impact alongside other training initiatives.
Finally, privacy and data handling will matter. Roleplay sessions involve sensitive workplace contexts. Even if the content is fictionalized, employees may still share personal or organizational details during practice. Enterprises will expect clear controls around data retention, access, and governance.
While the launch announcement emphasizes the product capabilities—live roleplay, feedback, scoring, analytics—the real-world success will depend on how confidently organizations can operationalize those capabilities.
The broader implication: AI training is becoming operational
Synthesia’s move reflects a larger shift in how AI is being used in corporate environments. Early AI training tools often focused on content generation: create videos, create scripts, create learning materials. Those tools reduced production costs,