Responsible AI, in practice

Mistakes are the point of practice, so the detailed record of a session, the conversation, the recording, the character's feedback, the debrief and the coaching chat, lives in the learner's private space. The organization sees how the program is going, not how one conversation went. In the training configuration, nothing a learner says in a session is used for evaluation, selection or promotion.

Scenarios, competency grids and feedback criteria are written and approved by people, yours and ours. The system does not learn from sessions or change its criteria on its own: every change is a human decision, released as a new version, so everyone is measured the same way over time. No decision about a person is automated, by Skillgym or by anyone using its data. See AI Ethics.

Feedback is built on observable behaviors recognized in the learner's own words, graded for the moment of the conversation, and aggregated on the client's competency model. The learner can see which sentence produced which assessment and why. The character's account of the conversation describes the character's reactions as the scenario determined them; nothing about the learner's emotions, voice or face is inferred.

Audio is transcribed and kept only so the learner can replay their own conversation, then deleted automatically after a period the client chooses. Before any content reaches an external AI service, direct identifiers are removed; the services work under contractual no-training clauses. Client data is never sold, is shared only with the processors listed in the Data Processing Agreement, and is never used to train any model. Details in Data Privacy Compliance and in the Privacy Policy.

Within that perimeter, the learner always holds the detail and the organization always holds the aggregate, you decide retention periods, who administers the program and what flows to your systems. Every user can access, correct or delete their data; the client can request deletion at any time, and all data is removed thirty days after the end of the contract. Requests through your privacy contact or our data request page.

Each client runs in a dedicated, isolated environment on AWS infrastructure, primary region in the EU, with encryption in transit and at rest. Other regions are available on request to meet data-residency requirements. Technical support is based in Switzerland. See Security Compliance.

Learners rate their experience; program owners review results with us; every calibration of a scenario or a grid is agreed with the client and documented. Because the system is non-self-learning, improvement is deliberate: nothing changes silently, and what changed is always traceable to a release.