AI Skills Taxonomy Architecture Director — Investment Banking

The mandate

The bank has licensed generative and predictive tools faster than it can distinguish foundational literacy from role-specific competence. Existing learning catalogues count attendance, job families describe yesterday’s work, and vendor badges do not prove judgement in regulated decisions. Leaders need a defensible view of which skills are required, evidenced, perishable and prohibited across materially different roles.

The engagement deliverable is an AI Workforce Skills Architecture linking work activities, risk tier, human accountability, knowledge, practical capability, evidence and renewal period. It must cover product, research, sales, operations, engineering, control and executive roles without implying that tool access equals proficiency. The artefacts will include assessment blueprints, progression routes, equivalence rules and governance ownership.

Five milestones govern delivery: week three accepts a work-and-risk inventory; week seven approves taxonomy and proficiency language; week twelve completes eight role-family pilots; week seventeen validates assessments against observed work; and week twenty-two accepts the architecture, adoption backlog and two transfer rehearsals. Fees follow written acceptance by People and Model Risk.

Acceptance requires business leaders to map twelve unseen tasks to required capability and human accountability, assessment owners to distinguish knowledge from demonstrated performance, and Compliance to reperform prohibited-use controls. The Chief People Officer signs only after internal teams update a role pathway for an unannounced model change without consultant interpretation or vendor-owned scoring logic.

The client will provide role profiles, work inventories, model and tool registers, risk classifications, learning records, assessment data, policy constraints, workforce plans, employee-representative agreements and named owners. The consultant does not select AI vendors, certify models, make employment decisions, design regulated controls, deliver mass training, negotiate with employee bodies or operate the skills platform.

Why this is external work

Technology vendors describe product features, business schools sell broad literacy and internal role owners naturally protect current structures. Independent skills architecture can join actual work, regulated accountability and observable evidence without promoting a platform, assessing named employees or deciding which roles should be removed.

What you will own

• Decompose AI-affected work into decisions, tasks, human accountabilities, risk conditions and observable performance evidence.
• Define literacy, practitioner, reviewer and accountable-owner proficiency without relying on course completion or self-rating.
• Map role families to required, optional, prohibited and expiring capabilities across front office and control functions.
• Build assessment blueprints using cases, work samples, challenge exercises, supervised practice and renewal evidence.
• Establish equivalence for prior experience, external credentials, internal academies and demonstrated workplace competence.
• Rehearse a new model release, changed policy, vendor-badge claim, role redesign and expired high-risk skill.
• Deliver taxonomy, pathways, assessment specifications, governance charters and a sequenced implementation backlog.

Candidate qualifications

• Designed enterprise skills taxonomies for AI, data or digital work across a large regulated financial institution.
• Converted changing work activities into proficiency evidence rather than broad competency labels or learning attendance.
• Built assessment and renewal models for practitioners, reviewers and accountable executives in high-risk roles.
• Connected role architecture, workforce planning, learning investment and internal mobility without automating employment decisions.
• Worked with model risk, compliance, employee representatives and technology leaders across clear decision boundaries.
• Transferred taxonomy governance to internal owners through live role changes and unfamiliar capability cases.

Non-negotiables

• Can lead all London role laboratories and two controlled taxonomy-change rehearsals within five months.
• Brings regulated AI workforce architecture at task level; generic digital-learning strategy is insufficient.
• Will disclose relationships with assessment platforms, AI vendors, training providers and labour-market data suppliers.
• Will not assess named employees, select vendors, certify models, decide redundancies or negotiate workforce agreements.

1. 49 words maximum. Describe an AI capability that course completion failed to prove in regulated work.
2. 49 words maximum. How would you separate practitioner skill from the accountability retained by a human reviewer?
3. 49 words maximum. Which model-change event would you use to test internal taxonomy ownership?