Data & Privacy

• Account details — name, work email, role, and the property/department you belong to.
• Roleplay transcripts — the text of what was said during a practice session, by both the learner and the AI character.
• Assessment results — a score and level per rubric criterion, the reasoning behind each, and the quote used as evidence.
• Delivery signals — timing-derived measures of a session: speaking pace, response time, share of talking, interruptions, and filler-word rate.
• Voice delivery scores — for communication-skills scenarios, how the session sounded (warmth, confidence, steadiness and similar), with the moment in the recording each observation refers to. These are coaching notes and do not count toward the overall score.

Maestro does not keep a copy of your session audio. It does not store video: the camera preview shown during a roleplay is for realism only — no video is recorded, transmitted, or scored. Maestro does not collect location data, contacts, or any personal information beyond the account details above.

Voice roleplays run on ElevenLabs, which records and retains the session audio on its own systems. Maestro stores only a reference to that recording, not the audio itself. That audio is currently retained indefinitely and is not automatically deleted. Deleting a learner from Maestro removes their Maestro data and requests deletion of their recordings from ElevenLabs — see “Deleting a learner’s data”.

• The learner — their own sessions, transcripts, and reports, unless their organisation has chosen to share results through managers instead (a per-organisation setting; learners can still request their data — see below).
• Their manager — learners in that manager’s property/department only. Managers cannot see other departments.
• Organisation admins — all learners within their own organisation.
• The Maestro platform team — for support and for operating the service.

• ElevenLabs — runs the voice conversation and holds the audio recording.
• Anthropic (Claude) — session transcripts are sent for grading against the rubric, and for tracking whether session objectives were met.
• Google (Gemini) — for communication-skills scenarios, a copy of the session recording is sent to score voice delivery, then discarded. Maestro does not retain the audio it sends.
• Neon — the hosted database where accounts, transcripts, and results live.
• Vercel — hosts and runs the application.

The Maestro database is hosted in the United States (AWS us-west-2), and the application runs in the United States (Vercel iad1). Sub-processors listed above may process data in other regions under their own terms. If your organisation requires data residency in a specific country, raise it before a pilot — it affects hosting choices.

Accounts, transcripts, and results are retained for as long as the organisation’s account is active, and are deleted on request. There is no automatic expiry today. Session audio held by ElevenLabs is retained indefinitely unless deletion is requested.

An admin or the learner’s manager can delete a learner from the People screen. This permanently removes their account, every roleplay transcript, all assessment results, and their assignments, and requests deletion of their session recordings from ElevenLabs. It cannot be undone. A learner who wants their data removed should ask their manager or organisation admin. An admin can also export everything held about a learner as a JSON file before deleting.

Maestro is a coaching and practice tool. Scores are generated by an AI model against a defined rubric and are intended to guide coaching conversations — every score carries the evidence behind it so a human can check it. Delivery signals such as speaking pace and filler rate are shown as context and are deliberately excluded from scoring, because they vary legitimately with accent, language, and personal style. Results should not be treated as the sole basis for an employment decision.