Privacy Policy
Last updated: October 29, 2026
Coachbench is a workbench for professional coaches: client management, engagements, sessions, workshops, assessments and the coaching tools around them. This page explains what data we store, why, for how long, who else sees it, and how you can export or delete it.
What data we store
Each row is a data category with its underlying tables (technical evidence), its purpose, the legal basis, the retention period, and its recipients.
- Account
- Your name, e-mail address and a hashed password (or a linked Google account) — Better Auth tables `user` / `account` / `verification`.
- Purpose: Sign-in and account recovery.
- Legal basis: Performance of the contract between the coach and Coachbench (Art. 6(1)(b) GDPR).
- Retention: Until you delete your account (Settings → Privacy). Deletion removes the account row immediately; sessions and linked OAuth accounts cascade with it.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Organization
- Your workspace: name, logo, optional custom document branding, and the subscription plan/status (`organizations`, `organization_branding`, `subscriptions`).
- Purpose: Operating the workspace you subscribed to; billing.
- Legal basis: Performance of the contract between the coach and Coachbench (Art. 6(1)(b) GDPR).
- Retention: Until the organization is deleted (sole owner deletes their account, or explicitly deletes the organization).
- Recipients: Hosting operator; a payment processor only once billing is actually enabled (no full card numbers are ever stored by Coachbench).
- Members and invitations
- Who belongs to your organization and at what role, plus pending seat invitations by e-mail (`memberships`, `org_invitations`).
- Purpose: Team access control.
- Legal basis: Performance of the contract between the coach and Coachbench (Art. 6(1)(b) GDPR).
- Retention: Memberships until removed or the account/organization is deleted. Invitations until accepted, revoked or expired.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Outbound e-mail (invitations, reminders, password reset) is sent through the provider the operator has configured (SMTP or Resend) — a sub-processor only when mail is configured at all.
- Clients
- Master data of the people or organizations a coach works with: name, company, e-mail, phone, address, notes, tags, status, AI-processing consent, and the manual communication log (`clients`, `client_interactions`).
- Purpose: Running the coaching relationship (CRM).
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Until the coach deletes the client record. Deletion cascades to contacts, teams, communication log, engagements, the originating lead (if the client was converted from one) and — per the deletion preview shown before every delete — the personal data listed there; see “Your rights” below.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Only reaches an AI provider when the coach actively uses a KI-Copilot feature AND — for anything client-scoped — the client's AI-processing consent is switched on; see “AI providers” below.
- Leads
- Prospects sent in by the coach's own marketing system through the organization's bridge key, before they become clients: name, company, e-mail, phone, source, campaign reference, notes and stage (`leads`). Only a one-way hash of the bridge key is stored (`bridge_keys`).
- Purpose: Following up on enquiries and turning them into client relationships.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Until the coach deletes the lead, deletes the client it was converted into, or the organization is deleted.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Contacts
- Named individuals at a client (sponsor, buyer, team lead, …): name, role, e-mail, phone, position, notes (`client_contacts`, `client_team_members`).
- Purpose: Knowing who to address within a client organization.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted together with the client they belong to.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Teams
- A client's team roster grouping contacts (`client_teams`), used by team pulses and assessment campaigns.
- Purpose: Measuring the same group repeatably (e.g. before/after a workshop series).
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted together with the client they belong to.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Engagements
- The container for a client relationship's lifecycle and light internal economics: title, objective, status, dates, contracted sessions, fee model/amount, notes (`engagements`).
- Purpose: Tracking a coaching engagement end to end.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted together with the client they belong to (cascade already in place).
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Contracts
- Coaching agreements — bilateral coach↔client or triangle coach↔client↔sponsor — as template-driven form content (`contracts`).
- Purpose: Documenting what was agreed (scope, confidentiality, reporting rules).
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted together with the client the contract is filed under; a contract with no client link is deleted with its project.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Sessions (incl. coach's private notes)
- Coaching session logs: template-driven content and the coach's CONFIDENTIAL private notes. Never included in any PDF/report; excluded by default from the client dossier export too, with an explicit, coach-only opt-in for the coach's own use (`sessions`, `session_templates` is the seeded, content-free template catalogue).
- Purpose: The coaching record itself.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted — including the coach's private notes — together with the client the session is filed under (previously such sessions merely lost their client link and survived; that gap is closed).
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Only reaches an AI provider when the coach actively uses a KI-Copilot feature AND — for anything client-scoped — the client's AI-processing consent is switched on; see “AI providers” below.
- Projects, activities and canvases
- Coaching/facilitation engagements across every tool (Design Thinking, Problem Solving, Leadership) with their phase progress and the content you type into method canvases and worksheets (`projects`, `project_phases`, `activities`, plus the Problem-Solving working tables `hypotheses`, `evidence_entries`, `ps_experiments`, `pdca_cycles`, `kata_storyboards`/`kata_steps`, `measures`, `accepted_causes`, `learn_logs`). Portfolio epic hypotheses (Scaling Workbench) reuse the same `hypotheses`/`evidence_entries`/`ps_experiments` tables, tagged `tool: portfolio_agility` — unlike the rest of this row, an epic hypothesis has NO project at all; it exists directly under the organization.
- Purpose: The actual coaching/facilitation work product.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Until the coach deletes the project (deletes everything listed here for that project) or, for a client-linked project, until the client is deleted (the project itself is kept and only unlinked; its content stays with the coach's own workspace). A portfolio epic hypothesis has no project to hang off: the coach deletes it directly on its own page (which also deletes its MVP experiment and evidence entries), otherwise it is kept until the organization itself is deleted.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Only reaches an AI provider when the coach actively uses a KI-Copilot feature AND — for anything client-scoped — the client's AI-processing consent is switched on; see “AI providers” below. Only with the organisation's opt-in for weekly tips: the NAMES of a member's own recently updated projects (never their content, never self-development projects, never projects of a client without AI consent) are sent to the AI provider once a week.
- Workshops, photos and participation
- Workshop agendas, decisions, actions, parking-lot entries and series (`workshops`, `workshop_items`, `workshop_decisions`, `workshop_actions`, `workshop_parking_entries`, `workshop_series`); uploaded flip-chart/room PHOTOS, which may show identifiable people (`workshop_photos` — for a photo flagged as showing identifiable people the uploader must confirm both storage and that the people shown consented to the photo being taken and used, and that confirmation is stored on the photo row with its time and the confirming account: `consent_confirmed_at`, `consent_confirmed_by_id`); and account-free participant input from live tools (dot voting, gradients, brainwriting, premortem) that is stored WITHOUT any identity, IP address or user agent by database design (`participation_links`/`participation_responses`, `participation_sessions`/`polls`/`poll_votes`/`submissions`).
- Purpose: Running and documenting facilitated workshops.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted together with the workshop; photo bytes are erased explicitly (not only the metadata row) when an organization or account is deleted. A client link on a workshop is removed (not the workshop) when that client is deleted. The consent record of a photo lives and dies with that photo — it is proof about the photo, so it is never kept once the photo is gone, and never dropped while the photo remains.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Only reaches an AI provider when the coach actively uses a KI-Copilot feature AND — for anything client-scoped — the client's AI-processing consent is switched on; see “AI providers” below.
- Assessments, campaigns and responses
- One measurement occasion per subject (`assessment_campaigns`), tokenised rater invitations that never link back to a response (`assessment_invitations`), and the submitted answers themselves — attributable only to a rater CATEGORY, never a person, and released only once a minimum-response floor is met (`assessment_responses`). The instrument questions themselves (`instruments`/`instrument_items`) are shared platform content, not personal data.
- Purpose: 360°/self-assessment, feedback collection, Character-OS profiling.
- Legal basis: Consent of the rater/assessed person, given at the tokenised invitation (Art. 6(1)(a) GDPR); processed on behalf of the coach for anything client-scoped.
- Retention: Deleted together with the client the campaign is filed under, or when the coach deletes it directly; a rater may also withdraw their own response via the one-time withdrawal link in their invitation.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Team pulses
- A thin before/after pairing of two team-pulse assessment campaigns for the same team (`team_pulses`); the actual responses live in the Assessments category above.
- Purpose: Showing the delta between two measurement occasions.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: The pairing record's client/team link is removed when that client is deleted; the pairing itself is kept as an org-level history record (its campaigns are deleted with the client per the Assessments row above).
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Sponsor objectives, three-way checkpoints and pulse responses
- Objectives the sponsor names for an engagement (`engagement_objectives`); the sponsor's own tokenised intake, never re-usable once submitted (`sponsor_intake_tokens`); the two contracted three-way checkpoints and the coach's own curated, releasable progress text for each objective — never a raw excerpt (`impact_checkpoints`, `impact_checkpoint_entries`); and the perceived-change pulse, whose responses carry a stakeholder CATEGORY (sponsor/manager/peer/direct report/other) but never an identity and only NUMBERS, never free text (`change_pulses`, `change_pulse_invitations`, `change_pulse_responses`). A task-only Character-OS blueprint (`character_os_task_blueprints`) also lives here — by construction it rates a TASK, never a person, and holds no personal data.
- Purpose: Proving coaching value to the sponsor within an agreed, curated boundary.
- Legal basis: Consent of the sponsor/stakeholder at the tokenised intake, where the legal basis and retention promised at that moment are snapshotted on the token itself (Art. 6(1)(a) GDPR); sponsors and stakeholders are named third parties who never consented to the coaching relationship as such.
- Retention: Deleted together with the engagement, which is deleted together with its client. Independently of that, the retention window promised at the tokenised intake is now actually enforced by a scheduled sweep: the sponsor's own entries (intake token, non-adopted objectives, the free-text criterion and confidentiality boundary of adopted ones, and the sponsor's own rating in a three-way checkpoint) are erased once the engagement has ended plus the promised window (default 12 months); stakeholder pulse answers and invitations are erased on their own clock, the promised window after the answer was submitted. An adopted objective keeps its title, because from the agreement onwards it is engagement content the coach's own records hang on.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Only reaches an AI provider when the coach actively uses a KI-Copilot feature AND — for anything client-scoped — the client's AI-processing consent is switched on; see “AI providers” below.
- Decisions
- Versioned decision records: frame, options, criteria, assumptions, dissent, the chosen option and rationale, plus later outcome reviews (`decisions`, `decision_reviews`).
- Purpose: A durable, auditable decision trail for the engagement.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: A decision is a durable audit record: deleting its project or client REMOVES the link only, it never deletes the record itself — you can delete a decision directly instead.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Only reaches an AI provider when the coach actively uses a KI-Copilot feature AND — for anything client-scoped — the client's AI-processing consent is switched on; see “AI providers” below.
- Boards
- Kanban and matrix boards for team, programme, large-solution and portfolio flow (the Scaling Workbench, M1-E40): board instances, their columns, matrix rows, cards (title, description, size label) and dependency arrows between cards, plus an append-only log of every column move (`boards`, `board_columns`, `board_rows`, `board_cards`, `board_card_edges`, `board_card_transitions`). The board template catalogue itself (`board_templates`, `board_template_columns`) is shared platform content, not personal data.
- Purpose: Visualizing and moving work through the team/programme/portfolio flow stages of the Scaling Workbench.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted when the coach deletes the board (cascades to its columns, rows, cards, dependency arrows and movement log), or when the organization is deleted. For a client-linked board, deleting the client REMOVES the link only (`boards.client_id` is set to null) — the board itself is kept and its content stays with the coach's own workspace, the same as a client-linked project or workshop.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Archived cases
- A closed Problem-Solving engagement's durable memory: copied title, tags, methods used, and free-text cause/solution summaries — deliberately DENORMALIZED so the record survives the deletion of the project or learn log it came from (`case_archives`).
- Purpose: Organizational memory and case-similarity hints for future engagements.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: When the client is deleted the client LINK is removed (anonymized) — the archive row itself is kept as organizational memory. This does NOT automatically scrub the client's name or other identifying detail from the free-text summary/tag fields if the coach typed one in; edit or delete the archive entry directly for that.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- Development goals, journals and nudges
- The Development Engine: goals, if-then/WOOP experiments, GAS ratings, Immunity-to-Change maps (`goals`, `experiments`, `gas_ratings`, `immunity_maps`); the client's OWN reflection journal, reachable via a capability-token link and structurally excluded from every sponsor/report surface (`journal_entries`, `journal_tokens`); and scheduled transfer/review e-mail nudges with their unsubscribe list (`transfer_nudges`, `nudge_suppressions`). Seeded program templates (`program_templates`) are platform content, not personal data.
- Purpose: Longitudinal development tracking and transfer support between sessions.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted together with the project it hangs off; a project stays with the coach's own workspace when its client is deleted, so this data is not touched by a client deletion unless the coach also deletes the project.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- AI audit metadata
- WHO called WHICH AI feature, WHEN, with which model (requested and actually used) and how many tokens — plus, in both `ai_invocations` and `copilot_audit_logs`, the cost of the call, the feature it served and whose API key paid for it. Deliberately content-free: neither table has a prompt or response column (`ai_invocations`, `copilot_audit_logs`).
- Purpose: Cost transparency, abuse detection, EU-AI-Act accountability.
- Legal basis: Legitimate interest in operating and securing the AI features responsibly (Art. 6(1)(f) GDPR).
- Retention: Until the organization is deleted; not user-deletable individually (it is the accountability record itself).
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient.
- AI key, budget and routing settings
- Which key pays for the organisation's AI calls and, if the organisation brings its own provider key, that key encrypted (never in plain text), its last four characters, when it was set and the user id of whoever last changed the setting (`ai_org_credentials`); the organisation's AI budget with its encrypted gateway key, period and spend baseline (`ai_org_budgets`); which budget threshold notices were already sent for a period (`ai_budget_notices`, no user id); purchased extra AI usage packs with pack, amount, period and the Stripe event id (`ai_extra_usage_grants`); on Enterprise, the organisation's own model choice per AI feature with the user id of whoever changed it (`ai_routes`). No prompt or AI output is stored here.
- Purpose: Running AI calls on the right key and model, enforcing and reporting the AI budget, and crediting purchased packs.
- Legal basis: Performance of the contract between the coach and Coachbench (Art. 6(1)(b) GDPR).
- Retention: Until the organization is deleted (all five tables are deleted with it). Removing the own key clears the key, its last four characters and its date; resetting a model choice deletes that row. Sent notices and purchased packs are kept until the organization is deleted. A user id stored as 'changed by' is not cleared when that user deletes their account. Not part of the organization export: the encrypted keys never leave the server; the key mode, the last four characters of an own key, the budget and the model choices are shown in the app's AI settings.
- Recipients: The hosting operator (see “Where your data is stored” above). An own provider key additionally goes, decrypted, to its own AI provider with each AI call of the organisation, to authenticate that call; nothing else in this category leaves the server.
- AI sparring conversations
- Conversations a coach holds with the AI sparring agent on a client page: the coach's own messages and the agent's replies, stored with that client and visible only to the coach who started them (`sparring_conversations`). Data the agent reads through its tools (decision outcomes, assessment deltas) is not stored again.
- Purpose: Letting the coach continue a preparation conversation across visits.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted automatically 90 days after the conversation was started. The coach can delete a conversation at any time; it is also deleted together with its client, with the coach's account and with the organization. When the client's AI consent is withdrawn, all conversations about that client are deleted at once; consent given again brings none back. When a coach leaves the organization or is removed, the conversations they held in it are deleted at once.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Only reaches an AI provider when the coach actively uses a KI-Copilot feature AND — for anything client-scoped — the client's AI-processing consent is switched on; see “AI providers” below.
- Hub agent proposals
- Changes the agent of a hub proposes to a coach in the guided interface, from fixed rules or from the coach's question: title, explanation, the change, a drafted text if any, the action's parameters, the prior values kept for undo, the ids of the clients concerned and its status (`agent_proposals`). Visible only to the coach it was made for. Nothing is ever sent to a client.
- Purpose: Letting the coach confirm a proposed change, and undo it afterwards.
- Legal basis: Processed on behalf of the coach, who is the data controller for their own clients (Art. 28 GDPR processor relationship).
- Retention: Deleted automatically 90 days after the proposal was made. Deleted at once when the AI consent of a client it concerns is withdrawn, when such a client is deleted, and when the coach leaves the organization or is removed; also deleted with the coach's account and with the organization.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. Only reaches an AI provider when the coach actively uses a KI-Copilot feature AND — for anything client-scoped — the client's AI-processing consent is switched on; see “AI providers” below.
- Recurring AI agents (weekly tips)
- Only on Team and Enterprise plans, and only once an owner or admin switches weekly tips on: the job setting with its schedule, last run and whether tips may use the organisation's own AI key (`agent_jobs`), the 1-3 tips each member receives in the app (`agent_tips`), and the organisation's permission record for reading typed content: who granted it and when, who revoked it and when (`org_agent_consents`). Without that permission, tips are built only from the content-free usage log (which feature, which action, how often). With it, the project names of the member's own recently updated projects are added; nothing else the member typed, never self-development projects, never projects of a client without AI consent.
- Purpose: Practical tips for daily work, written once a week.
- Legal basis: Performance of the contract: an optional feature the organisation switches on (Art. 6(1)(b) GDPR). Project names are read only after the organisation's explicit, revocable permission.
- Retention: Tips are deleted after 90 days, with the member's account, or with the organisation. When a member leaves the organisation or is removed, their tips in it are deleted at once. Permission records are kept until the organisation is deleted, as evidence of when reading was allowed. Revoking the permission stops reading at the next weekly run; switching tips off stops the runs. Tips already delivered stay until their 90 days are up, also after tips are switched off or the permission is revoked.
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. On Team and Enterprise plans with weekly tips switched on, the AI provider (see “AI providers” below) also receives each member's own content-free usage counts once a week. An organisation on its own AI key gets tips only after an owner or admin allows it (off by default); they then go to that organisation's own provider on its own key. Only with the organisation's opt-in for weekly tips: the NAMES of a member's own recently updated projects (never their content, never self-development projects, never projects of a client without AI consent) are sent to the AI provider once a week.
- Technical session data
- Login sessions needed to keep you signed in, including IP address and browser user-agent (Better Auth `session` table); a content-free, best-effort adoption log of which feature was used, with no free text or names (`usage_events`).
- Purpose: Keeping you signed in; product-usage insight for the operator.
- Legal basis: Performance of the contract (staying signed in, Art. 6(1)(b) GDPR) and legitimate interest in security and product operation (Art. 6(1)(f) GDPR).
- Retention: Sessions until sign-out or expiry, and in any case until the account is deleted (cascade).
- Recipients: The hosting operator only (see “Where your data is stored” above); no other recipient. On Team and Enterprise plans with weekly tips switched on, the AI provider (see “AI providers” below) also receives each member's own content-free usage counts once a week. An organisation on its own AI key gets tips only after an owner or admin allows it (off by default); they then go to that organisation's own provider on its own key.
Where your data is stored
Coachbench runs on a server rented from Hetzner Online GmbH, Industriestr. 25, 91710 Gunzenhausen, Germany. The server and its database are located in Hetzner's data centre in Nuremberg, Germany (EU); backups are also stored in Germany. We do not sell your data or share it with third parties for advertising.
AI providers
AI features run server-side. Features that work with a client's data need that client's explicit consent (see "Your rights" below) and send only allowlist-checked fields. Some features also send text the coach typed, not only such fields: for example, the Editorial Coach sends the problem description and the answers the coach types into its dialogue, and client-prep sparring sends the messages the coach types into the conversation. Weekly tips send each member's own content-free usage counts (which feature, which action, how often) without any per-client consent. Weekly tips add the names of the member's own projects only if an owner or admin of the organization has allowed it, and never the projects of a client without AI consent. Each row shows what an AI feature actually sends and what is kept afterward.
Hub agents (guided interface) work with data of clients who consented to AI, and with data that belongs to no client. Items linked to a client without AI consent are left out. In addition, before any request, text that names such a client is removed: the full name, and the surname or company name where it is unambiguous (not a first name alone, not a short or common word, not a surname a consenting client or a contact of one also has). A hub's suggestions come from fixed rules and send nothing to a model. Only "Ask the agent" sends your question and short facts from that hub to the model, using your organization's AI budget or its own provider key, like sparring. A question that names such a client is refused; otherwise your question is sent as you wrote it, so please do not type details of clients without consent into it. An agent changes data only after you confirm. Most changes can be undone right after they are made. Undo is refused when the data has changed since; drafts change nothing and need no undo. Nothing is ever sent to a client: drafted texts are previews you send yourself, and clear requests to send something are refused without a model call.
Editorial Coach
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
Only the problem description and the answers the coach typed into the Socratic dialogue are sent. No client master data, no CRM fields.
No prompt or model output is stored by the platform. Only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp) is retained for accountability and cost reporting. The provider does not train on API data.
Problem Solving — AI copilot
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
Allowlisted fields only (problem statement, cause labels, decision frame, revealed premortem reasons). Participant names and identities are never sent; premortem reasons are read only after the round is closed and revealed.
No prompt or model output is stored by the platform. Only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp) is retained for accountability and cost reporting. The provider does not train on API data.
Leadership — AI copilot
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
A minimized, org-scoped project context (goals, GAS ratings, recent protocol excerpts). Journal entries are never fetched; pattern analysis additionally requires a per-project opt-in.
No prompt or model output is stored by the platform. Only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp) is retained for accountability and cost reporting. The provider does not train on API data.
Facilitation — AI copilot
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
Coach-view data only: the facilitator's own agenda and observations, plus a server-injected method catalogue. No participant data and no anonymous-vote content reach the model.
No prompt or model output is stored by the platform. Only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp) is retained for accountability and cost reporting. The provider does not train on API data.
Client-prep brief
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
Allowlisted fields from exactly two sources: decision-audit outcomes (title, status, open concerns) and assessment score deltas (dimension keys and numeric means). Reflection-journal content and raw session free-text are never read; deltas appear only above the anonymity floor; the client's identity is never sent.
No prompt or model output is stored by the platform. Only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp) is retained for accountability and cost reporting. The provider does not train on API data.
Sponsor report — wording assistance
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
Only the already-released sponsor report model: the contracted objective titles, the coach's own curated progress text, and min-n-suppressed aggregate means. Session content, journal entries and individual stakeholder answers are never read, so they cannot be sent; suppressed figures reach the model as nulls, never as numbers.
No prompt or model output is stored by the platform. Only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp) is retained for accountability and cost reporting. The provider does not train on API data.
Client-prep sparring
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
The coach's own messages in the conversation, plus, only when the agent asks for them through its two read tools, the same allowlisted fields as the pre-session brief: decision-audit outcomes (title, status, open concerns) and assessment score deltas above the anonymity floor. Reflection-journal content, session free-text and the client's identity are never read.
The conversation (the coach's messages and the agent's replies) is stored with the client so the coach can continue it. The coach can delete it at any time; it is deleted automatically 90 days after it was started, and together with the client, the coach's account or the organization. When the client's AI consent is withdrawn, it is deleted at once; consent given again brings none back. When the coach leaves the organization or is removed, it is deleted at once too. Data read by the agent's tools is not stored. No prompt or model output is stored in the audit log, which keeps only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp). The provider does not train on API data.
Weekly tips
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
Runs once a week without anyone present. By default only the member's own content-free usage counts of the last 7 days are sent (which feature, which action, how often). Only after an owner or admin of the organisation has opted in, the names of the member's own projects updated in that week are added (at most 10, never self-development projects, never projects of a client without AI consent). Nothing else the member typed is read.
The tips themselves are kept in the member's in-app inbox for 90 days, also after tips are switched off or the permission to read project names is revoked. When a member leaves the organization or is removed, their tips in it are deleted at once. No prompt or model output is stored in the audit log, which keeps only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp). The provider does not train on API data.
Hub agents (guided interface)
Anthropic PBC (EU-Standardvertragsklauseln / EU SCCs)
Only when the coach asks the agent of a hub a question: the question as typed, plus short read-only facts from that hub (data of clients with AI consent and data that belongs to no client), each with its source. A fact linked to a client without AI consent is removed before the request, and such a client is never the subject of a proposal. In addition, a fact or source naming such a client or one of its contacts is removed, and a question naming one is refused before any request: by full or normalised name, or by surname or company name when that is a distinctive word; never by a first name alone, and not by a short or common word or a surname a consenting client or a contact of one also has. Otherwise the question is sent as written. The agent cannot send anything. Clear requests to send, invite or forward are answered without a model call. The hub's suggestions without a question are rule-based and send nothing to the model.
The answer is not stored. Proposals (from rules or from a question: title, explanation, the change and any drafted text) are stored for the coach they were made for, so they can be confirmed and, unless the data has changed since, undone; nothing is ever sent to a client. They are deleted automatically 90 days after they were made, at once when the AI consent of any client they concern is withdrawn or such a client is deleted, at once when the coach leaves the organization or is removed, and together with the coach's account or the organization. On a consent withdrawal, any other proposal whose text names that client by name (not by first name alone) is deleted too, unless it was already applied: then it keeps its change and undo data, and only its text is replaced. No prompt or model output is stored in the audit log, which keeps only content-free metadata (org, user, feature, model, token counts, cost, key source, timestamp). The provider does not train on API data.
Transfer to Anthropic PBC (USA) under the EU Standard Contractual Clauses (SCCs). An organization can instead use its own provider key (Settings → AI key). Its AI requests then go to the provider under that organization's own account: the terms with the provider, including the basis for the transfer, are the organization's own agreement with that provider. Coachbench stores such a key encrypted and passes it to the provider with each request. Weekly tips are not produced for an organization on its own key.
Why we process your data
Two different roles, not one: for account, organization and billing data, Coachbench is itself the controller (performance of the contract with you as a coach, Art. 6(1)(b) GDPR). For your clients' data — master data, sessions, workshops, assessments and everything else you record about them — the coach is the controller, and Coachbench processes it on the coach's behalf (Art. 28 GDPR). The precise legal basis per category is in the data catalog above.
How long we keep it
We keep your data for as long as your account and organization exist, with these exceptions. Client-prep sparring conversations are deleted 90 days after they were started; the coach who started one can delete it earlier at any time. When a client's AI consent is withdrawn, all sparring conversations about that client are deleted at once; consent given again brings none back. Weekly tips are deleted from the member's inbox after 90 days; they stay that long also after the organization switches tips off or revokes the permission to read project names. When a member leaves the organization or is removed, their weekly tips and the sparring conversations they held in it are deleted at once. A sponsor's own entries (the intake, objectives not adopted, the success criterion and confidentiality boundary of adopted ones, and the sponsor's ratings in three-way checkpoints) and pulse invitations are erased once the engagement has ended plus the retention window promised at the intake (12 months by default); an adopted objective keeps its title. Stakeholder pulse answers are erased the promised window after they were submitted, independently of the engagement. Leads that your own marketing system sends in through your organization's bridge key are kept until you delete the lead, delete the client it was converted into, or delete the organization. Deleting your account also deletes your sparring conversations and your weekly tips. Deleting a client deletes their sessions (incl. the coach's private notes), contracts, sponsor objectives and responses, three-way checkpoints, pulse responses, contacts, teams, communication log entries, campaigns with their responses, the sparring conversations about them, and the lead it was converted from (if any); archived cases are anonymized (the client link is removed, free text is left as-is unless edited separately); projects are kept and only unlinked. The preview shown before every deletion states the exact counts. Backups, where they exist, are rotated on a limited schedule and are not separately scrubbed.
Hub agent proposals (guided interface) are deleted 90 days after they were made. When a client's AI consent is withdrawn, every proposal concerning that client is deleted at once. Any other proposal whose text names that client by name (not by first name alone) is deleted too, unless it was already applied: then it keeps its change and its undo data, and only its text is replaced. When a member leaves the organization or is removed, their proposals are deleted at once. Deleting your account or a client deletes the proposals that belong to it.
Your rights: access, rectification, export and deletion
Access and portability: from Settings → Privacy you can export a complete JSON copy of your organization at any time; every client page also offers a complete client dossier export. Rectification: all master data can be edited directly in its forms. Deletion: you can delete a single client (with a preview of what is deleted, what is anonymized, and what is only unlinked), delete your organization entirely, or delete your whole account. Deletion is immediate and cannot be undone. Withdrawing AI consent: the toggle on every client page. If you are in a region with statutory data rights (e.g. GDPR access, rectification, erasure, restriction and portability), these controls are how you exercise them; contact us for anything the self-service tools do not cover — in particular a request from a client who is not themselves a user of the platform: please forward it to your coach, who responds as the controller.
Restriction of processing (Art. 18 GDPR) and objection to processing (Art. 21 GDPR) are deliberately not exposed as switches in the application: please assert these two rights with your coach as the controller. They decide on them and arrange for them to be carried out; the operator acts only on their instruction in this respect. If your request concerns your own account and your own organization, use the contact given below instead.
Contact
Operator of this Coachbench instance and contact for privacy requests: Tim Zeitzen, freelance consultant for agile management, Westermoor 7a, 28844 Weyhe, Germany. Phone: +49 155 67403574. E-mail: contact@agile-management.eu. Please use this address for any privacy question the in-app tools do not cover.