Role-Play Coach Agent: AI Skills Practice and Evaluation
Simulates real conversations to train skills, and evaluates performance in real time. The Role-Play Coach Agent (also known as Gi Practice) is part of Gi's team of four AI Agents working 24/7. Your organisation's data stays in your organisation.
What it does
The Role-Play Coach Agent creates interactive simulations where employees practise difficult conversations (feedback, sales, leadership) with an AI counterpart. It analyses performance throughout the conversation and delivers structured feedback at the end of the session.
Key capabilities
Interactive scenarios with an AI counterpart: not scripted branches, real dialogue
Feedback, sales, leadership, customer service: the most common high-stakes conversation types
Performance evaluation throughout the conversation: live signals, not just an end summary
Structured feedback at the end of the session: what worked, what to improve, what to try next
Use cases
Sales, leadership, feedback, customer service. Examples of who uses it:
Sales teams: discovery calls, objection handling, negotiation
Managers and team leaders: feedback conversations, performance discussions, difficult announcements
Customer service teams: escalations, complaint handling, de-escalation
HR and L&D: interview practice, onboarding conversations, compliance scenarios
A concrete example
Scenario: Giving difficult feedback.
The simulated employee says: "I feel my work isn't being recognised by the team."
The Role-Play Coach Agent replies: "Thank you for sharing. Can you give me a specific example from this week?"
Real-time metrics shown to the learner during the conversation include Empathy 82% and Clarity 76%, with the final feedback at the end of the session.
How it works
An admin or designer defines a scenario: the situation, the role the AI will play (for example, a sceptical customer or an underperforming team member), the goals of the conversation, and the success criteria.
The learner then engages in a natural conversation with the AI character. The AI adapts to what the learner says, pushes back, asks follow-up questions, and reacts the way the persona would react. There are no scripted branches to memorise: each session unfolds like a real exchange.
When the conversation ends, the agent produces a structured evaluation: what went well, what to improve, and concrete recommendations the learner can apply in the next attempt.
What the evaluation includes
The evaluation is built around the success criteria defined for the scenario, and typically covers:
Coverage of key points: did the learner address the topics the scenario required?
Tone and approach: was the conversation empathetic, clear, and professional?
Outcome: was the goal of the conversation reached?
Specific recommendations: concrete suggestions tied to moments in the transcript.
The evaluation is available to the learner straight after the session, and can be reviewed by managers and trainers in the backoffice.
What you can do in the backoffice
Design scenarios: define the role, situation, and evaluation criteria.
Publish to specific groups: scope a scenario to a department, a function, or a cohort.
Review individual sessions: open a learner's transcript and evaluation for coaching follow-up.
Track engagement: see who is practising, how often, and how their evaluations evolve.
Privacy and data handling
The Role-Play Coach Agent uses an external AI provider (OpenAI API, under a contractual no-training clause). The provider receives the scenario context, the conversation turns, and a pseudonymous identifier for the learner. It does not receive the learner's name, email, or manager name. The provider is contractually prevented from using prompts or responses to train its models.
All conversations and evaluations stay inside your GFoundry environment, which is hosted in the European Union. Retention follows the windows configured for your tenant. Employees can request deletion of their session history through their administrator.
What it does not do
It does not replace human coaching. The agent gives consistent, scalable practice; humans handle context, nuance, and growth conversations.
It does not evaluate personality. The output assesses the practised skill in the scenario, not the learner as a person.
It does not feed external systems. Sessions and evaluations stay within your GFoundry environment.
Tips for better practice
Make the scenario specific. "A customer asking for a 30% discount on a renewal" beats "a tough customer call".
Pair with a course. Use the Learning Designer Agent to build the underlying course, then the Role-Play Coach Agent as the practice step.
Review and iterate. After a few sessions, refine the success criteria based on what learners actually do.