People Analytics Model Risk Recovery Leader — Retail Banking

The mandate

Attrition, absence, skills and promotion models are embedded in manager dashboards, workforce plans and talent reviews, yet their inventories and decision impacts are incomplete. Several models were built as exploratory analyses and became operational without validation, drift thresholds or worker notice. The analytics head departed after Internal Audit found sensitive proxies and unrecorded local model variants.

The ten-month assignment starts within two weeks and covers model containment, three workforce decision cycles, independent validation proof and permanent-leader induction. A new model, feature, population, vendor version, threshold, use case, jurisdiction or adverse outcome becomes a governed risk event. Six weeks are reserved for handover; no extension will cover routine people analytics delivery.

Exit requires one model and analytic inventory, consequence tiers, intended-use boundaries, data lineage, validation evidence, fairness and stability tests, human review, monitoring, worker rights, retirement controls and three cycles within tolerance. The successor must adjudicate an unseen attrition-model drift affecting a protected cohort before accepting model-governance authority.

The interim may suspend unvalidated models, remove prohibited features, restrict dashboards, require owner attestations, direct remediation, replace temporary analytics leads and control HK$95 million of authorised recovery funding. Business leaders retain employment decisions; Privacy and Legal interpret obligations; independent validators challenge methodology; employee relations owns individual process and remedy.

Individual selection, performance rating, promotion, redundancy, legal advice, psychological assessment, disciplinary investigation and replacement of the analytics platform are outside scope. The leader may halt model use but cannot decide employee outcomes. Historical predictions will not be treated as ground truth, and accuracy gains will not excuse unexplained cohort harm or incompatible secondary use.

Why this seat is open

Exploratory workforce analysis quietly became decision infrastructure without the controls expected for consequential models, and the leader’s departure removed accountable challenge. Temporary authority must inventory real use, restore validation and human ownership through live decision cycles, then prove a successor can govern drift and cohort impact.

What you will own

• Inventory statistical models, scores, segments, rules and derived features used across workforce decisions and planning.
• Map intended purpose, actual users, decision influence, affected populations, jurisdictions and accountable human owners.
• Govern training data, lineage, proxy review, validation, stability, fairness, explainability and performance thresholds.
• Establish approval, use restriction, human review, monitoring, incident, challenge, correction and retirement controls.
• Compare central and local variants for undocumented feature, threshold, population and decision-purpose changes.
• Command scenarios involving drift, data outage, vendor update, new jurisdiction and adverse cohort outcome.
• Transfer three complete decision cycles and an unseen model-impact case to the permanent analytics risk leader.

Candidate qualifications

• Held people-analytics model governance authority in a bank, insurer or other highly regulated employer.
• Recovered predictive and segmentation models that moved from exploration into consequential workforce use.
• Evaluated lineage, validation, proxies, drift, fairness, explainability and human review across employee populations.
• Restricted unsafe analytics without taking promotion, performance, redundancy or individual employment decisions.
• Worked with independent validation, Privacy, Legal, employee relations and Internal Audit across clear boundaries.
• Handed recovered model governance to permanent leadership through repeated live decision cycles and complex cross-population cohort-impact scenarios.

Non-negotiables

• Available within two weeks for full-time Hong Kong leadership and three workforce-decision cycle observations.
• Direct consequential people-model governance experience is required; dashboard development alone is insufficient.
• Will disclose relationships with analytics platforms, assessment vendors, data providers and validation firms.
• Will suspend unvalidated use despite business demand and will not decide outcomes for named employees.

1. 49 words maximum. Describe an exploratory workforce model that became consequential without adequate governance.
2. 49 words maximum. How did you detect proxy harm that aggregate accuracy concealed?
3. 49 words maximum. Which drift event must the permanent leader adjudicate before handover?