Why AI Roleplay Alone Won’t Make Your Workforce Ready

NovoEd’s take on “Leveraging AI Role-Play to Build True Performance Readiness” – summary from article published in Training Industry, October 2026

Author: Farnaz Ronaghi, Co-founder and Chief Product and Technology Officer, NovoEd

The Short Version

Our co-founder Farnaz Ronaghi recently wrote for Training Industry about a question we hear from nearly every L&D leader evaluating AI role-play: what does it actually take for practice to change performance?

Her starting point is a gap AI has made impossible to ignore. Anyone can now ask a chatbot how to handle a pricing objection or structure a hard conversation and get a workable answer in seconds. But knowing the answer doesn’t make someone steady when the conversation goes sideways. Readiness takes knowledge plus the confidence and context to use it under pressure.

In the article, she makes the case that AI role-play paired with human coaching is one of the most important shifts available to L&D right now, and she’s equally direct about why many deployments stall. We’ve summarized her argument below and added where NovoEd stands. For the full piece, including her five evaluation questions, read it on Training Industry.

A Few Terms From the Article:

• Performance readiness: Knowledge plus the confidence and context to act on it in a real work situation, under pressure.
• The practice bottleneck: Practice has long been known to build skill, but live role-play depends on available experts and is hard to scale, and scripted branching simulations are expensive to build and quickly memorized.
• Practice as an island: When a role-play tool sits apart from the learning program, its feedback never reaches the coach, manager, or L&D team who could act on it.
• Practice data: Evidence of how people respond in simulated situations and where they get stuck, as opposed to completion data, which only shows who finished.

What Farnaz Argues in the Article

1. The evidence for practice has been there for decades. She points to a 2005 Journal of Applied Psychology meta-analysis of behavior modeling training (Source: Taylor, Russ-Eft, and Chan), which found that skill and on-the-job behavior gains held up over time even as knowledge gains faded. Transfer was strongest when goal setting and supervisor involvement were part of the design. The problem was never whether practice works. It was making practice available to more than a select few.
2. Generative AI removes the scripting constraint. Scenarios no longer need every branch written in advance, so learners can practice the same skill repeatedly without having the same conversation twice, and get immediate feedback against a defined rubric in the organization’s own language.
3. AI changes what human coaches get to work on. When a manager arrives at a coaching session having already practiced, the coach can see where they struggled and start there. Human time shifts to the situations a rubric can’t settle, like a burned-out high performer or a conflict with years of history.
4. Deployments stall on relevance and disconnection. Ready-made scenario libraries are fast to launch, but employees take them less seriously when the situations don’t resemble their actual work. And when practice lives in a separate tool, L&D ends up maintaining the same frameworks in several places while the results go nowhere.
5. “Embedded vs. standalone” is the wrong question. Farnaz argues that a standalone tool can be well integrated and a platform feature can still be poorly designed. Instead, she offers five questions for evaluating any approach, covering grounding, learning objectives, maintenance effort, targeted practice, and what happens with the results. The full list is in the article.
6. Scale is not the same as impact. She’s careful to separate reaching more people from proving performance improved. The more important question is what organizations do with the practice data once thousands of people start generating it.

How We See Things at NovoEd

1. Practice lives inside the learning experience. Learning, AI Roleplay, and 1:1 coaching run in one system, so what happens in practice informs what happens in coaching.
2. Scenarios are built from your own content. Feedback is scored against your organization’s methodology, not a generic standard.
3. Practice goes beyond role-play. Learners can respond to scenarios, record their screen while using job tools, or submit written work, each with immediate feedback.
4. Managers see where their people need help. Practice results surface for the managers and coaches who can act on them.

Frequently Asked Questions

What is AI role-play in corporate training?
AI role-play uses generative AI to simulate realistic workplace conversations, such as a sales objection or a performance conversation, so employees can practice and get immediate feedback without a live facilitator.

Does AI role-play replace human coaching?
No. As Farnaz argues in Training Industry, AI provides baseline practice at scale so that coaches, managers, and experts can focus on harder situations and on judgment that goes beyond a rubric.

Why do AI role-play deployments fail?
The two most common reasons are relevance, when scenarios don’t reflect the organization’s real work, and disconnection, when practice results never reach the coaches or programs that could act on them.

What is the difference between practice data and completion data?
Completion data shows who finished a course. Practice data shows how people performed in simulated situations, which behaviors they demonstrated, and where they struggled.

How should L&D evaluate AI role-play tools?
Look at whether practice is grounded in your organization’s frameworks, tied to explicit learning objectives, easy to keep current, able to target specific weaknesses, and connected to what happens next. Farnaz’s full set of evaluation questions is in her Training Industry article.