Skip to main content
This document answers frequently asked questions about AI usage at Jam Technologies GmbH. It is intended to support internal approval processes and discussions with stakeholders such as IT departments and works councils. For additional questions, contact clemens@wejam.ai.

​
How Jam’s AI System Works

Jam’s platform supports human teams through AI-powered conversation simulations (voice-based role-plays with fictional personas) and improvement-oriented personalized feedback. The system operates in four phases:
  1. Content creation - Training administrators create scenarios, personas, and scorecards. Customer information flows into AI prompts only as instructions, is visible and editable at any time, and is never used for model training.
  2. Conversation simulation - The employee conducts a voice-to-voice role-play with an LLM-powered AI persona in real-time.
  3. AI evaluation - A separate LLM analyzes the transcript against the scorecard criteria and generates improvement-oriented feedback.
  4. Results - The employee receives the audio, transcript, and AI feedback. Coaches and supervisors can access results only if the customer explicitly enables this.
AI is used for speech recognition/transcription, conversation simulation, feedback evaluation, and as a content-creation copilot for training administrators. Key data protection facts:
  • All data is stored in the EU
  • Employees see their own data and evaluations
  • DPAs with all AI providers; no use of customer data for AI training
  • Supervisor access to training data can be individually granted or denied

​
Does Jam Train AI Models on Customer Data?

No. Jam uses pre-trained models from third-party providers (primarily OpenAI). Customer data - uploaded documents, audio, transcripts - is never used to train the underlying AI models. The only way customer information enters the AI is through prompts: instructions for conversation simulations and transcript analysis. These prompts are under full control of the customer’s training administrators and can be viewed and modified at any time in Jam Studio. OpenAI’s Data Processing Agreement (DPA) explicitly states that data transmitted via the API is processed solely for service delivery and not used to improve or train OpenAI’s models. Jam will provide the relevant DPA passages to your data protection team on request.

​
Is the AI System Autonomous or Self-Learning?

No. Jam deliberately avoids autonomous self-learning. All AI behavior is controlled through human-authored prompts - the role-play persona and scenario, and the feedback scorecard with its evaluation criteria. Any changes after go-live are made by human training administrators via the CMS. There is no automated individual decision-making about users as defined by Art. 22 GDPR.

​
Risk Classification

Jam classifies the risk of its AI system as low, based on:
  • Sole purpose: training employee conversation skills
  • No sensitive personal data (gender, ethnicity, health data) in the system
  • No high-stakes autonomous decisions about users (e.g. hiring decisions)
  • Data stored encrypted on European servers under GDPR
  • Full customer control over prompt content
This assessment has been verified using the EU AI Act Compliance Checker at ai-act-service-desk.ec.europa.eu.

​
Quality Monitoring and Emergency Procedures

Jam monitors AI feedback quality continuously through Sentry and Betterstack for real-time anomaly detection, regular manual reviews of AI outputs, customer feedback tracking via Jira Service Desk, and periodic bias checks across user groups. If quality issues arise, Jam can shut down AI feedback functionality immediately by deactivating services and revoking API credentials. Affected customers are notified transparently. All incidents are documented in the incident management process. AI feedback accuracy is >93% compared to human expert annotations of the same transcripts. Only 0.26% of all AI-generated feedback has received negative user ratings via the in-app thumbs-down button.

​
Data Protection & User Rights

Where is data stored? All personal data is stored in European data centers - primarily Heroku PostgreSQL databases in AWS EU regions and Amazon S3 in the EU. How long is data retained? Based on contract term and legal retention obligations. After contract termination or upon written request, Jam deletes all customer data per Section 12 of its DPA. After contract end, data is available for 14 days for download before permanent deletion. Are users informed when AI evaluates them? Users are the first and primary recipients of all AI output. Feedback is improvement-oriented and based solely on the role-play transcript. No conclusions about a user’s person in the broader sense are drawn. Can users object to AI feedback? Yes. Users can rate AI feedback as unfair via the in-app thumbs-down button at any time. Hints of bias or unfair evaluations are escalated immediately and lead to technical adjustments. Does Jam have published terms of use? Jam concludes individual SaaS contracts including a DPA with each business customer, which contractually regulate all usage and data protection aspects. These documents are available on request for review by your data protection officer.

​
Contact

Questions about AI usage, the DPA, or internal approval documentation: Dr. Clemens Lechner - Chief Product Officer
clemens@wejam.ai