Compliance — FirstLineCoach

The architectural principle

AI never participates in preparing a coaching session with a specific individual.

The tool draws a clear line between two moments. At session time, when a coach searches for plans to prepare for a session with a specific person, everything is deterministic — no AI. At publication time, when content is being created or published, AI can help — but no specific individual is being assessed. An AI system that influences which coaching approach a specific person receives would fall within scope of Annex III, Point 4(b) of the EU AI Act as a potentially high-risk system. That is why the separation is deliberate.

Coaching session preparation

Plan search is fully deterministic: text normalisation, suffix stemming, and token-set matching. No AI, no inference, no probabilistic model.
Excluded from the AI Act's scope by Recital 12: rules defined by humans, executed by traditional software.
The loop never closes back to the person you're coaching.

Plan design and publication

AI assists during vignette capture in /reflect — to help you draft your reflection. You review, edit, and decide whether to save.
When a plan is published, AI generates search vocabulary from the plan content, reviewed by a human editor before it enters the search index.
AI compares vignettes against existing plans during triage — an editorial tool, not a person assessment.
Coaching signals are extracted anonymously — no link to the source vignette, quarter-only timestamps.
Every vignette is human-reviewed before it can be adopted as a coaching plan.
The voice transcript is deleted immediately when you save. It is never persisted to disk.
You can delete any of your vignettes at any time.

EU AI Act (Regulation (EU) 2024/1689)

The tool is limited-risk under the Act's classification. It is not used for employment decisions, performs no biometric categorisation, and runs no emotion recognition in the sense prohibited by Art. 5(1)(f).

Annex III, 4(b): Covers AI systems intended to allocate tasks based on individual behaviour in employment relationships. This is why plan search at session time is deliberately deterministic.
Art. 50(1): The /reflect screen makes clear that your voice is transcribed and sent to an AI (Anthropic Claude).
Art. 50(4): Vignettes are AI-drafted content under human editorial control — you review and save.
Art. 4: AI literacy — this page explains in plain language what the AI does, what it does not do, and why.
Art. 5(1)(f): The tool does not infer the emotions of the person you're coaching. You describe what you observed; the AI only reorganises your own description.
Recital 12: Plan search is traditional rule-based software — not an AI system.

GDPR (Regulation (EU) 2016/679)

Art. 5(1)(c): Data minimisation — only the vignette is kept. The transcript is discarded on save. Coaching signals are anonymous.
Art. 5(1)(e): Vignettes are kept until you delete them. Transcripts are not stored at all.
Art. 6(1)(a): We process your reflection with your consent. Pressing Save is the consent.
Art. 6(1)(f): Legitimate interest covers incidental mentions of third parties in your captured speech. The transcript is discarded on save — no personal data from your speech is retained.
Art. 13: This page is the transparency notice.
Art. 17: You can delete any vignette at any time. Deletion is immediate and removes both language versions.
Art. 22: No automated decision is made about you or about the person you're coaching.

Data controller

Ennovate Consulting, Ireland. Contact: cmurphy@ennovateconsulting.ie.

Hosting: Microsoft Azure, West Europe (Ireland). AI processor: Anthropic, under their Data Processing Addendum and EU Standard Contractual Clauses.