Human Capital CXO
Legal & Trust — ETH-RSS-001
Ethics & Data Governance Framework
Effective: 2026-06-01  •  Version 1.2  •  Review cycle: Annual
GDPR Art. 22 Compliant Swiss nDSG 2023 Human Oversight Required No Automated Adverse Decisions EU AI Act — High-Risk Category
Jump to Responsible Use Bias Controls Calibration Rules Privacy by Design Legal Framework FAQ
🚫
Absolute Prohibition — Automated Adverse Employment Decisions
The E(RVBER) output must never be the sole or automated basis for termination, demotion, pay reduction, redundancy selection, or any other adverse employment action. Every use affecting an individual requires a qualified human decision-maker, corroborating evidence, documented rationale, and an employee right-of-appeal. This prohibition is non-negotiable and applies to all deployments, integrations, and API consumers.
Pillar 1

Responsible Use Policy

The HCV Model is a decision-support tool — not a decision-making system. It produces an indicative monetary estimate of human capital value to support strategic HR analysis, M&A due diligence, and workforce planning. All outputs must be interpreted by a qualified professional.

✓ Permitted Uses
  • M&A due diligence — valuing human capital of a target organisation
  • Strategic workforce planning and HR budgeting
  • Organisational design analysis and restructuring modelling
  • Compensation benchmarking (informing, not determining, pay)
  • HR analytics research, internal reporting, and board presentations
  • ERP / SAP integration for HC reporting dashboards
  • Academic research and educational demonstration
  • Expert witness valuation in employment or transaction disputes
✗ Prohibited Uses
  • Automated or sole-basis termination of employment
  • Automated pay reduction without human review and employee notification
  • Automated demotion, role reassignment, or redundancy selection
  • Hiring rejection based solely on predicted future E(RVBER)
  • Systematic scoring of employees using protected characteristics as inputs
  • Real-time surveillance or continuous automated monitoring of individual value
  • Profiling for loan, insurance, or credit decisions
  • Any deployment without disclosure to the affected employee or works council
👤
Human Oversight Requirement
Every use of E(RVBER) that affects a named individual requires review by a qualified HR professional or people manager before any action is taken. The reviewer must document their independent assessment alongside the model output.
🔔
Employee Notification
Employees must be informed when their personal data is being processed to compute an HCV valuation, consistent with GDPR Art. 13/14 and nDSG Art. 19. The purpose, legal basis, and data categories used must be disclosed.
⚖️
Right of Appeal
Any employee who is subject to a decision informed by HCV Model output has the right to contest that decision, request a human review, and receive an explanation of the factors used. This mirrors GDPR Art. 22(3) safeguards.
📋
Documentation Obligation
Organisations using the HCV Model for decisions affecting employees must maintain a record of: inputs used, model version, date, the human reviewer's name, their independent assessment, and the final decision rationale. Minimum retention: 5 years.
Pillar 2

Bias Risk Controls

The dual control factor architecture — EI(cf) and IM(cf) — introduces dimensions that could encode demographic bias if not carefully defined and validated. The following controls are mandatory for any professional deployment.

Factor / Dimension Bias Risk Risk Level Required Control
EI(cf) — Market Skill Demand Demand proxies may correlate with age, gender, or ethnicity if based on job titles rather than verified competencies Medium Must reference validated external labour market data (e.g. BFS, LinkedIn Talent Insights). May not use demographic proxies.
EI(cf) — External Career Alternatives Assuming fewer alternatives for older workers, women, or non-majority ethnic groups encodes discriminatory assumptions High Score must be based on documented, role-specific market data. Demographic characteristics are prohibited inputs.
EI(cf) — Professional Reputation Social visibility and external networks correlate with gender and ethnicity; "reputation" scores can amplify systemic inequalities High Score must be based on documented, observable outputs (publications, client feedback, project results) — not social media presence or network size.
IM(cf) — Culture & Environment Fit "Culture fit" is a well-documented proxy for demographic homogeneity and has been scrutinised in employment discrimination case law High Must be replaced by or grounded in "values alignment" defined against documented, objective criteria. Requires legal review before use in any employment decision.
IM(cf) — Peer Leadership Peer assessments may encode in-group favouritism along demographic lines Medium Peer leadership scores must be validated against org-wide distributions. Scores that differ by >15% across demographic groups trigger mandatory recalibration.
IM(cf) — Subordinate Satisfaction Satisfaction surveys reflect power dynamics and may disadvantage managers from minority groups who face hostile team environments Medium Use validated psychometric instruments. Disaggregate results and review for demographic patterns before including in IM(cf) scoring.
Service State Definitions Defining "Exceeds Expectations" or "Marginal Performer" without objective criteria embeds evaluator bias High All service states must be defined using observable, documented performance criteria reviewed by HR and legal. Annual validation required.
Mobility Probability Matrix Historical attrition and promotion data reflects past discrimination; using it uncritically perpetuates it Medium Audit mobility data for demographic bias before use. Apply corrective adjustments if promotion or attrition rates differ significantly by group.
Pillar 3

Calibration Rules

Accurate E(RVBER) output requires rigorous, documented calibration of every input parameter. The following rules are mandatory for professional deployments.

1

Service State Definition — documented & legally reviewed

Each service state (Exceeds Expectations, Meets Expectations, Marginal Performer, Exit) must be defined in writing using observable, objective criteria. Definitions must be reviewed by HR and employment law counsel before use.

Minimum: 3 documented, observable criteria per service state. Annual review required.
2

Service State Values — market-anchored compensation data

Reward values expressed as % of base salary must reference current market compensation data (e.g. Mercer, Kienbaum, Swiss wage statistics). Surrogate measures must be documented and justified.

Update at least annually or whenever compensation structures change materially (>10% shift).
3

Expected Service Life — based on role-specific data

The time horizon n must reflect the realistic tenure expectation for the specific role and organisation, not a generic assumption. Reference industry attrition data and internal historical tenure where available.

Use minimum 24 months of internal historical data where available. Cap horizon at 7 years unless role-specific data supports longer.
4

Mobility Probability Matrix — validated, bias-audited

Transition probabilities between service states must be derived from actual historical data, audited for demographic bias, and signed off by an HR professional. Expert estimation is permitted where data is insufficient, but must be flagged as such in the output.

Recalibrate within 30 days of any significant organisational change (restructuring, M&A, mass hiring event).
5

EI(cf) Scoring — anchored to observable market criteria

Each external individual sub-factor must be scored against documented, specific criteria. Scores must not embed assumptions about protected characteristics. Cross-validate scores against market data quarterly.

Score anchors must be documented in a calibration register. Scores deviating >0.20 from peer group average require documented justification.
6

IM(cf) Scoring — Likert instrument validated annually

Internal managerial control factors must use validated psychometric instruments aligned with Likert's organisational measurement dimensions. Scoring rubrics must be applied consistently across the organisation and audited for inter-rater reliability.

Cronbach's α ≥ 0.70 required for any IM(cf) survey instrument. Annual recalibration with cross-group validation.
7

Discount Rate — independently reviewed

The discount rate r must reflect the organisation's current weighted average cost of capital (WACC) or a defensible sector benchmark. It must be reviewed by finance and documented with its source and date.

Review discount rate at least annually. Document source (e.g. "Swiss National Bank base rate + 3% risk premium, Q1 2026").
8

Cross-Group Validation — demographic neutrality test

Before any deployment affecting employment decisions, the calibrated model must be tested for systematic output variance across demographic groups. Results that show >10% mean difference by gender, age band, or ethnicity require investigation and correction before use.

Document validation results in calibration register. Retain for 5 years minimum.
Pillar 4

Privacy-by-Design Architecture

The HCV Model processes personal data (compensation, performance state, tenure) that is sensitive in an employment context. The following architecture principles apply to all deployments under Swiss nDSG 2023 and GDPR.

📥
Data Collection
Minimise: collect only annual base salary, service state assignment, probability estimates, EI(cf) and IM(cf) sub-scores, and time horizon. No health, family status, or other sensitive data.
⚙️
Processing
Pseudonymise: replace employee names with internal IDs before entering the calculation engine. The mapping table (ID ↔ name) is held separately with restricted access.
🗄️
Storage
Retain minimally: individual valuations: 24 months maximum. M&A transaction records: 10 years (legal obligation). Anonymised aggregate data: indefinite. All data encrypted at rest (AES-256).
🔑
Access Control
Need-to-know only: individual valuations accessible only to authorised HR staff and deal team members. Role-based access control. All access logged with timestamp and user ID.
🗑️
Erasure
Right to erasure: employees may request deletion of their valuation data. Process within 30 days. Exception: data required for legal proceedings or regulatory obligation.
📤
Data Subject Rights
Full rights: right to information (GDPR Art. 13/14, nDSG Art. 19), right of access (Art. 15), right to rectification (Art. 16), right to object (Art. 21), and right to human review of automated processing (GDPR Art. 22 / nDSG Art. 21).
GDPR Article 22 — Automated Individual Decision-Making

Article 22 of the GDPR grants data subjects the right not to be subject to a decision based solely on automated processing where it produces legal or similarly significant effects. HCV Model deployments must implement the following safeguards to comply:

  1. A qualified human decision-maker reviews the E(RVBER) output before any employment action
  2. The employee is informed that their data was processed by an algorithmic model
  3. The employee is provided a meaningful explanation of the factors and their weightings
  4. The employee has the right to contest the output and request re-evaluation
  5. The decision-maker's identity, assessment, and rationale are documented separately from the model output
  6. A data protection impact assessment (DPIA) is completed for any systematic employee valuation programme
Questions

Frequently Asked

Can the model be used to decide who to make redundant? ▾

No — not as a standalone or automated input. E(RVBER) can inform a redundancy analysis by providing an indicative monetary value of each role's human capital, but the selection decision must be made by a human manager following your jurisdiction's redundancy process, equality legislation, and consultation requirements. The model output must never be the sole criterion for selection.

Does integrating the model with SAP create additional compliance obligations? ▾

Yes. Embedding the HCV Model in an ERP system that continuously processes employee data likely triggers a DPIA requirement under GDPR Art. 35 and nDSG Art. 22. You should also review whether the integration constitutes a high-risk AI system under the EU AI Act — if so, conformity assessment and registration (Art. 49) become mandatory. Works council consultation may be required in Germany, Austria, and other co-determination jurisdictions.

Is "culture fit" a safe input to the IM(cf) factor? ▾

Only with caution and legal review. "Culture fit" has been challenged in discrimination case law as a proxy for demographic homogeneity. If used, it must be grounded in specific, documented, objective criteria (e.g. "alignment with documented values of transparency and client focus") rather than subjective assessments of personal style or background. We recommend replacing "culture fit" with "values alignment" and grounding each criterion in observable, documented behaviours.

Does the employee have a right to see their E(RVBER) valuation? ▾

Under GDPR Art. 15 and nDSG Art. 25, employees generally have a right of access to personal data held about them, including processed outputs derived from their personal data. Where E(RVBER) is computed using identified employee data, the individual may request access. The response should include: the output value, the input parameters used, the model version, the date, and contact information for the data controller. Seek legal advice before establishing your subject access request process.

Is the model classified as high-risk under the EU AI Act? ▾

Yes, for employment-affecting deployments in the EU. The EU AI Act 2024/1689 entered full application in August 2026. Annex III classifies AI systems used for "recruitment or selection of natural persons, promotion and termination of work-related contractual relationships, task allocation, monitoring or evaluation of performance and behaviour of persons in work-related contractual relationships" as high-risk. Conformity assessment (Art. 43), technical documentation (Art. 11), transparency (Art. 13), human oversight (Art. 14), and EU AI database registration (Art. 49) are legally mandatory — not optional. Designate an EU authorised representative under Art. 22 where required before EU commercial deployment.

What is required for M&A use — is employee consent needed? ▾

In M&A due diligence, processing employee data is typically based on legitimate interests (GDPR Art. 6(1)(f)) or legal obligation rather than individual consent. However, the target company must still satisfy transparency obligations, data minimisation principles, and — in most jurisdictions — notify the relevant works council or employee representatives. Individual employees are generally not entitled to block the due diligence process, but they retain their rights of access and rectification for post-transaction data.

Download the Formal Policy Document
The standalone Responsible Use Policy is available as a clean, printable document suitable for legal review, enterprise procurement, and works council documentation.
View Policy Document → Compliance enquiry →