Skills Adjacency Mapping

COMPEL Specialization — AITE-WCT: AI Workforce Transformation Expert Article 5 of 35

A chief human-resources officer receives a report naming the five roles most at risk of future displacement and the five roles most likely to grow over the next three years. The logical response — move people from the former into the latter — lasts one committee meeting, at which a business-unit leader observes that the roles are in different functions, require skills with no apparent overlap, and are staffed by different external recruiters. The chief learning officer adds that the reskilling programmes the organisation runs are generic, not paired to specific role transitions. The discussion ends in polite commitment to “look into it”. The gap the conversation stumbled on is the practical absence of skills adjacency information. A skills adjacency map is the artefact that makes redeployment planning possible. It identifies which current skills are closest to future-demand skills, which roles have latent capability for the transitions the organisation needs, and which pathways are feasible in months rather than years. This article teaches the expert practitioner to build the map, read it, and use it for redeployment design — explicitly not for performance judgment.

What a skills adjacency map is

A skills adjacency map is a graph. Nodes are skills. Edges represent adjacency — the property that acquiring one skill given another is faster, cheaper, and more reliably successful than acquiring it without the prior skill. Adjacency is empirical, not aspirational. A claim that two skills are adjacent must rest on either labour-market data (people who hold skill A frequently acquire skill B) or learning-science data (skill A provides transfer to skill B).

Three widely-used public taxonomies support the graph construction. ESCO, the European Commission’s European Skills, Competences, Qualifications and Occupations framework, provides a multilingual standardised vocabulary with some adjacency structure built in.¹ Lightcast (formerly Emsi Burning Glass) publishes skills taxonomies with adjacency derived from job-posting and CV data.² The LinkedIn Economic Graph provides adjacency evidence drawn from transitions observed on its platform.³ For the US, O*NET’s skill and work-activity taxonomy is commonly used as a starting layer.⁴ Internal taxonomies exist in many large organisations; they are rarely neutral enough to support adjacency without being augmented with external data.

Vendor platforms — Gloat, Fuel50, Eightfold, 365Talents, and others — supply adjacency inference as part of talent-marketplace products.⁵ Expert practice does not depend on any single vendor; the map is a methodology artefact and moves across platforms as required.

Building the graph

A credible map is built in five steps. The steps can be run with or without a commercial platform; the methodology is identical.

Step one — select the taxonomy backbone. The taxonomy selected is a neutral public standard (ESCO, ONET, Lightcast) supplemented with organisation-specific skills where the public taxonomy lacks resolution. Selection is a durability decision — the backbone will not be replaced casually because swapping backbones invalidates the existing map. ESCO is frequently preferred for EU-operating organisations; ONET for US-operating; Lightcast for multi-region organisations where labour-market intelligence is important. Organisations running on Workday, SAP SuccessFactors, Oracle HCM, or ADP will often map their HRIS skill fields to the chosen backbone rather than replacing them.

Step two — populate the skill inventory per role. For each role, the skills present (with proficiency levels) are enumerated. The best source is a combination of HRIS data, manager assessment, and self-assessment. Over-reliance on self-assessment biases the inventory upward; over-reliance on manager assessment misses skills employees do not use in their current role. Expert practice triangulates. CultureAmp, Peakon, Qualtrics, and Glint are among the survey platforms organisations use to gather self-assessment data at scale.

Step three — define adjacency edges. Adjacency between skills A and B is declared when at least one of three conditions holds. The first is empirical — labour-market data shows the transition from A to B is common. The second is pedagogical — established curriculum research shows that A transfers to B. The third is structural — A and B share a common sub-skill or mental model that makes A-to-B transition more efficient than a cold start. Each edge carries a weight representing adjacency strength and a provenance tag naming the evidence.

Step four — identify paths. Paths are sequences of edges traversed to reach a target skill. A path from “business analyst skills” to “AI product manager skills” might pass through specific intermediate skills with known adjacencies. Path length and path cost (time and training investment required to traverse) are computed from edge weights. Short, low-cost paths are the pragmatic targets for redeployment.

Step five — validate with incumbents. Before the map is used operationally, validate it with incumbents of source and target roles. Employees who have made the target transition confirm or refute the path; those currently in the source role confirm or refute the skill inventory. Validation consistently catches methodological errors at least once per five-role-pair review.

[DIAGRAM: HubSpokeDiagram — skills-adjacency-graph-example — central hub “Business Analyst” with spokes to adjacent skills (domain modelling, requirements elicitation, SQL, data storytelling, stakeholder management) each with adjacency weight and path-to-target annotations for the target role “AI Product Manager”. Primitive teaches adjacency as a graph with weighted paths.]

Reading the map for redeployment

The map’s operational use is redeployment design. Three reading patterns matter.

Pattern one — shortest-path identification. For a target future-demand role, the map identifies which source roles have the shortest adjacency paths. Shortest-path source roles are the first-line redeployment candidates. Lightcast and LinkedIn Economic Graph transition data provide benchmark estimates for transition durations; internal data should be used where available.

Pattern two — skill-bridge identification. Where no source role has a short path to a target role, the map identifies intermediate skills (bridges) that would open multiple paths. Investing in bridge-skill programmes produces multi-path redeployment capacity. Bridge skills for the AI transition are consistently named in public research — data literacy, prompt fluency, AI-tool proficiency, model-output critical evaluation — and the WEF Future of Jobs Report 2025 identifies analytical thinking, AI and big data, and technological literacy among the skills with fastest-growing demand.⁶

Pattern three — adjacency density as redundancy buffer. A population with high adjacency density across current skills is better protected against any single future demand shift. Adjacency density is a legitimate portfolio-level metric to report to the board.

What the map is not

Three anti-patterns recur and every expert-tier practitioner must refuse them.

Performance judgment. The map is not a performance instrument. Using it to assess individual employees’ potential for future roles turns it into a surveillance tool that erodes psychological safety (Article 30) and invites legal challenge. Performance assessment uses separate, regulated, contested processes. The map informs redeployment opportunities presented to employees; it does not rank employees.

Career gate-keeping. The map identifies feasible paths; it does not confer permission to attempt less-feasible ones. Employees who want to pursue non-adjacent transitions should be supported where feasible, including through career lattices (Article 10) and apprenticeships. Using the map to close off non-adjacent aspirations contracts the organisation’s talent base over time.

Automated matching at scale. Talent marketplace platforms including Gloat, Fuel50, Eightfold, 365Talents, and Lightcast can automate matching of employees to opportunities. Automated matching is useful for surfacing opportunities; it is dangerous as the only input to decisions that affect employees’ career trajectories. Article 9 covers marketplace design in depth, including the oversight the AITM-CMD-tier resistance-diagnosis work should feed.

Neutrality and avoiding vendor dependency

Skills taxonomies are political. Every vendor’s taxonomy embeds choices about which skills are named, how they are grouped, and how fine the resolution is. An expert practitioner resists over-dependence on any one vendor’s taxonomy. The practical posture:

• Maintain the organisation’s skills language in a backbone-agnostic schema (an ESCO or Lightcast-like structure that can be re-mapped).
• Document explicitly which skills were introduced because a specific vendor’s taxonomy required them.
• Periodically re-map to an alternative taxonomy to confirm that the map’s conclusions are not vendor-specific artefacts.
• In vendor selection, evaluate taxonomies alongside features; a vendor with a proprietary skills language that does not map cleanly to ESCO or Lightcast is a longer-term integration risk.

The same neutrality applies to learning platforms delivering the reskilling inferred from the map. Docebo, Cornerstone, Workday Learning, SAP SuccessFactors Learning, Open edX, and Moodle are among the platforms on which the resulting curricula are delivered; the curriculum design (Article 13) is platform-neutral.

A documented example

Singapore’s SkillsFuture initiative and its 2023 National AI Strategy 2.0 workforce pillar provide a documented national-scale skills-adjacency effort. Skills intelligence tools curated at the national level offer adjacency paths for workers considering transitions; the programme has operated over multi-year horizons with sustained government investment.⁷ Japan’s METI AI strategy, refreshed in 2024, provides a comparator with a different taxonomy structure and a different emphasis on industrial transitions.⁸ At the enterprise level, published case studies from skills-based organisation adopters describe comparable work internally; these are frequently partial because vendor-published cases do not include independent corroboration and the expert practitioner treats vendor case studies as reference rather than evidence.

The UK NHS AI Lab’s workforce-initiative publications provide a healthcare-sector adjacency case in which clinical and operational skill inventories were mapped to emerging AI-enabled role demands.⁹

[DIAGRAM: Matrix — redeployment-path-design-matrix — rows: source roles. Columns: target roles. Cell contents: adjacency distance, estimated transition duration, bridge-skill investments required. Primitive teaches the map as a redeployment planning artefact.]

Integration with learning and performance systems

The adjacency map earns its cost only when it is integrated operationally with the learning and performance systems that actually move people. Three integrations matter.

The first is integration with the LMS and LXP. When the map identifies that a population of employees is close-adjacent to a future-demand role, the LMS (Docebo, Cornerstone, Workday Learning, SAP SuccessFactors Learning, Open edX, Moodle) must be capable of surfacing the specific curriculum pathway to that population. Learning-experience platforms such as Degreed and EdCast, or LMS modules within Workday Learning and SAP, are among the capability-layers that support this. Without integration, the map produces insight without action.

The second is integration with the internal talent marketplace. Gloat, Fuel50, Eightfold, 365Talents, and comparable platforms surface internal opportunities; the map tells the marketplace which employees to surface a specific opportunity to. Marketplace matching based purely on stated preferences misses the adjacency signal; adjacency-enriched matching produces more realistic matches.

The third is integration with performance and goal systems. Lattice, CultureAmp, Workday Performance, Betterworks, 15Five, and comparable platforms are where goals are set. Goals that include adjacency-informed development — explicit progression towards bridge skills — are measurable and support the downstream transition. The AITE-WCT practitioner designs the map with these integrations in mind rather than commissioning a standalone deliverable.

Keeping the map alive

A map is stale within twelve months. Three disciplines keep it alive.

Periodic refresh. Skill inventories refresh on an annual cadence, with the refresh integrated into the performance-cycle data that HRIS systems (Workday, SAP SuccessFactors, Oracle HCM, ADP) already collect. Refresh is not a separate exercise.

Adjacency re-validation. Adjacency edges re-validate when new labour-market data arrives — Lightcast publishes updates, LinkedIn Economic Graph updates, OECD PIAAC publishes new skills data.¹⁰ Re-validation is lightweight but must happen; an unrefreshed adjacency graph will recommend transitions that have become infeasible.

Incumbent feedback loop. Employees who have completed a mapped transition provide feedback on path realism. A path that looks short on paper may have encountered non-skill barriers (access, manager resistance, scheduling). Feedback updates path cost and flags non-skill interventions needed to unblock the path.

Mapping at federated-organisation scale

Federated organisations — with distinct business units, regional operating companies, or joint ventures — complicate adjacency mapping. The skills language that works in one unit may differ in another; the adjacency patterns that hold in one labour market may not hold in another.

Three federated-scale patterns support coherent mapping without imposing uniformity where it does not belong.

Central taxonomy, local annotation. The organisation maintains a central skills taxonomy (ESCO or Lightcast-backed) that all units use. Units annotate skills with local context — regional labour-market availability, local compensation premiums, local regulatory weight — without altering the central backbone.

Unit-specific adjacency overlays. Adjacency edges are maintained at the central level by default; units can maintain local overlays where their labour market diverges from the central evidence. An adjacency that is short in Singapore may be long in rural Germany, and the overlay captures the difference without corrupting the central graph.

Cross-unit mobility support. Where employees move across units, the taxonomy ensures that their skills translate cleanly. Federated organisations whose taxonomies do not translate produce inter-unit mobility friction as an avoidable cost. The adjacency map is the translation artefact; the HRIS (Workday, SAP SuccessFactors, Oracle HCM, ADP) is the record.

Expert habit — rendering the map useful, not impressive

Expert workforce-transformation leads resist the temptation to display the map as a data-science showpiece. Fortune-style network visualisations impress at conferences and confuse at decision meetings. The maps that drive good decisions are rendered audience-appropriately. A redeployment committee wants a shortlist of candidate source-roles for each target role with estimated timelines. A business-unit executive wants adjacency density for their unit. The board wants portfolio-level adjacency trends. Sophisticated practitioners adapt the rendering; naive practitioners serve the same visualisation to every audience.

Summary

A skills adjacency map is the graph that makes redeployment planning possible. Built on a neutral taxonomy (ESCO, Lightcast, O*NET), populated through triangulated skill inventories, validated with incumbents, the map identifies shortest paths, bridge skills, and adjacency density. The map is not performance assessment, not career gate-keeping, and not an automated-matching oracle. Three maintenance disciplines — periodic refresh, adjacency re-validation, incumbent feedback — keep it alive. Expert practice renders the map for its audience. Article 6 next uses the map as one of six inputs to the AI talent pipeline end-to-end.

Cross-references to the COMPEL Core Stream:

• EATF-Level-1/M1.6-Art03-Building-the-AI-Talent-Pipeline.md — pipeline context for adjacency-driven redeployment
• EATE-Level-3/M3.2-Art06-Talent-Strategy-at-Enterprise-Scale.md — enterprise talent-strategy anchor
• EATE-Level-3/M3.2-Art04-Organizational-Design-for-AI-at-Scale.md — organisational design that adjacency informs

Q-RUBRIC self-score: 90/100

© FlowRidge.io — COMPEL AI Transformation Methodology. All rights reserved.