AI Roleplay FAQ

Frequently Asked Questions: AI Roleplay

When users complete Roleplays in Showpad's Web App, they'll receive AI-generated feedback. Read more about AI Roleplays here.

To read more about general Generative-AI FAQs, continue here. 

Data Privacy & Security

Is my voice or Roleplay conversation used to train AI models? 

No. Your data is never used to train the AI models.

• External Models: We explicitly specify in our contracts with third-party LLM providers (Large Language Models) that your data is not to be used for training purposes.
• Internal Models: Customer data used in our internally hosted models is isolated; it is not accessible by other customers and is not used to train our base models.

Is my practice session recorded? 

Yes, but with strict controls on where it is stored.

• Processing (The "Thinking" Phase): When you speak, your audio is sent to our third-party AI providers solely to generate a response. These providers operate under a zero-data-retention policy, meaning your data is deleted immediately after processing.
• Storage (The "Review" Phase): The recording and transcript are stored securely within the [Company Name] Secure Cloud, not on the public AI model, so that you and your manager can review your performance later.
   

Who has access to my Roleplay data? 

Only authorized users within your organization.

• Access: Your Roleplay transcripts and audio are stored in our private, authenticated "Recording and Transcript Store". They are never exposed to the public internet or other customers.
• Processing: The "GenieAI" gateway acts as a secure intermediary, stripping personal identifiers before sending necessary text to the LLM for response generation.

Architecture & How It Works

How does the AI generate a voice response? 

The system uses a secure, multi-step process to ensure realism and speed without compromising privacy:
Speech-to-Text (STT): Your spoken audio is converted to text by a secure provider.

• Scenario Processing: The RolePlayAI engine sends this text (along with the scenario rules) to the LLM via GenieAI to generate a reply.
• Text-to-Speech (TTS): The text reply is converted back into audio using a specific "Voice Profile" (the persona of the buyer) and played back to you. Note: The "Voice Profile" determines how the AI sounds (e.g., "skeptical buyer"), it does not clone your voice.

What happens if I mention sensitive client info? 

While our system is secure, we recommend treating the AI as a Roleplay partner rather than a real sales call.

• Best Practice: Do not use real Personal Identifiable Information (PII) or specific proprietary secrets during a Roleplay.
• Safety Net: The system is designed to process the conversation contextually for scoring, not to memorize facts for a database.

Guardrails & Scoring

How is my performance evaluated? 

After your session, you can request an evaluation. This triggers the Evaluate Workflow:

1. RoleplayAI fetches the full transcript of your session from the secure store.
2. The transcript is analyzed by the LLM against the specific "Rubric" or "Success Criteria" defined for that scenario (e.g., "Did the rep ask open-ended questions?").
3. You receive a score and qualitative feedback based on that specific attempt.

Why did I get a different score for a similar answer? 

The AI scores are based on the conversation's context.

• Subjectivity: Just like a human manager, the AI evaluates tone, phrasing, and the flow of the conversation.
• Consistency: Unlike humans, the AI applies the same rubric every time. However, if your phrasing was slightly less confident or clear in one attempt, the score may reflect that nuance.