Job analysis

Job analysis is the upstream work-definition process behind many HR systems: selection, assessment, performance management, compensation, training, work design, and legal defensibility. It defines the tasks, KSAOs, work contexts, and performance domains that later measurement and decision systems treat as job relevant.

2022-strah-diversity-issues-job-analysis shows why job analysis should not be treated as automatically neutral just because it is systematic. If job-analysis data capture majority-group norms, biased archival language, unequal SME voice, or averaged ratings that erase subgroup differences, then downstream HR systems can inherit a narrow definition of successful performance. The foundation can be procedurally careful and still construct-deficient.

The paper also reframes job analysis as a diagnostic tool, not just a documentation tool. Done inclusively, job analysis can reveal whether different demographic groups are assigned objectively different tasks, face different expectations, use different successful performance strategies, or experience work constraints that the majority definition of the job misses.

For AI workforce transformation, job analysis matters because claims about automation, augmentation, and role redesign depend on knowing what work actually is. AI adoption and work-redesign arguments become weak when they rely on stale occupational descriptions or task lists that do not capture how people actually perform, adapt, and coordinate work.

2025-tomlinson-working-with-ai shows one AI-era update to that problem: use observed AI conversations to map user goals and AI actions back to ONET intermediate work activities, then aggregate those activities to occupations. The method keeps the job-analysis lineage visible, but it also shows the limit of static taxonomies: ONET enables cross-occupation comparison while potentially missing current task change, non-U.S. work, and the connective tissue between listed activities.

2008-icf-competency-modeling-job-analysis adds the competency-modeling boundary. It distinguishes task-oriented job analysis, worker-oriented job analysis, bottom-up competency modeling, and top-down organizational competency models. The useful distinction is not “job analysis versus competencies” in the abstract; it is whether the method is defining current tasks, future tasks, job-level KSAOs, cross-job competencies, or organization-level strategic capabilities.

2009-sanchez-levine-competency-modeling-job-analysis adds a purpose boundary. Traditional job analysis is strongest when the task is to describe current or past work, document job-relevant KSAOs, and support defensible HR decisions. It becomes weaker when leaders need to signal future-oriented strategic behavior across roles; that is where competency modeling can supplement job analysis, provided the broad themes are later translated into job-family behavioral indicators.

2015-neubert-assessment-21st-century-skills adds another boundary: job and work analysis can define current tasks and occupational requirements, but it may under-serve skills that matter precisely because work is becoming more nonroutine, dynamic, and collaborative. CPS and ColPS do not replace job analysis; they point to the kind of problem features, interaction demands, and adaptive knowledge use that job analysis has to represent before assessment can be job-relevant for changing work.

2025-teyssier-roberge-21st-century-skills-overlap adds a warning for AI-era job analysis and skills taxonomies: evergreen skill labels are attractive when work is changing quickly, but broad labels such as adaptability, collaboration, critical thinking, and self-management can overlap semantically and psychometrically unless the analysis specifies the work behaviors, problem contexts, and assessment evidence that make each skill job-relevant.

2026-putka-indexing-ai-impact-onet makes the job-analysis consequence concrete for AI-impact measurement. Its proposed O*NET index suite separates declarative knowledge, procedural knowledge and skill, task performance, contextual performance, and whole-job performance instead of treating a task list as the entire criterion domain; it also preserves a distinction between direct occupation-level judgments and occupation scores aggregated from lower-level elements.

Connections

• Relates to construct-validity because construct validity begins before measurement: the job and criterion domain have to be defined coherently and inclusively before predictors can be validated against them.
• Relates to competency-modeling because competency models can either be rigorous job-level KSAO maps or broader strategic/cultural models that require different evidence standards.
• Relates to competency-modeling because competency modeling can supply strategic direction that job analysis does not, while job analysis supplies the behavioral specificity and job-relatedness logic that broad competencies often lack.
• Relates to complex-collaborative-problem-solving because CPS and ColPS translate nonroutine and interactive work demands into assessable problem features that job analysis should not flatten into static task lists.
• Relates to algorithmic-assessment because algorithmic hiring and assessment tools inherit their target definitions from upstream work analysis; model validity cannot repair a deficient job definition.
• Relates to work-redesign because AI-mediated roles require renewed analysis of tasks, KSAOs, work contexts, and human-agent workflow boundaries.
• Relates to inclusive-hr-systems because job analysis is one mechanism by which inclusion either enters or fails to enter recruitment, selection, performance management, promotion, and compensation.
• Relates to evidence-based-management because HR evidence claims often rest on documented job relevance; the quality of that documentation is itself an evidence question.
• Relates to human-capital-resource-measurement because job analysis helps decide which skill evidence can be aggregated into workforce capability claims rather than merely counted as generic future-skill inventory.

Contradictions & open questions

• Job analysis is often treated as the safeguard that makes personnel systems fair and legally defensible. 2022-strah-diversity-issues-job-analysis complicates that assumption: the safeguard can transmit bias if it defines the job from majority-group experience.
• The field has long treated SME disagreement as possible error or unreliability. A central open question is when disagreement reflects noise versus real subgroup differences in task exposure, expectations, constraints, or ways of performing the work.
• AI-era work redesign raises a timing problem: job analysis can document current work, but agentic workflows may change tasks faster than traditional job-analysis cycles can update them.
• 2025-tomlinson-working-with-ai sharpens the timing problem: AI-use traces can update the map of work activities faster than traditional job analysis, but they may overrepresent what people ask chat tools to do and underrepresent embedded workflow changes, avoided uses, and accountable job context.
• 2008-icf-competency-modeling-job-analysis complicates the legal-defensibility instinct: strategic job analysis and competency modeling may be better suited to future work, but they may also be less empirically settled or less specific than traditional task analysis.
• 2009-sanchez-levine-competency-modeling-job-analysis complicates the incumbent-data default: if future role enactment is the target, incumbents may not be the right primary source, but leader-defined strategic competencies still need work-analysis methods before they become assessment or selection criteria.
• 2015-neubert-assessment-21st-century-skills complicates the current-task default: CPS and ColPS are attractive because they model dynamic problem interaction, but workplace use still needs evidence that the simulated problem features match the focal job, role, or redesigned workflow.
• 2025-teyssier-roberge-21st-century-skills-overlap complicates the evergreen-skills impulse: future-oriented skill taxonomies can become more durable than current task lists, but only if job analysis keeps them tied to observable behavior and avoids treating overlapping labels as distinct job requirements.
• 2026-putka-indexing-ai-impact-onet argues that contextual performance is systematically missing from task-centric AI-impact measures, while also proposing LLM-generated contextual behaviors where O*NET has no occupation-specific list. That creates a new upstream question: can generated behaviors represent the social and organizational performance domain without importing model stereotypes or flattening local work context?