How to analyze workforce skills with AI

Define the work before the skill

Start with future and current tasks, decisions, tools, constraints, and expected outputs. A label such as “AI skills” is too broad to guide investment. Describe observable capability: for example, “verify claims in an AI-assisted client brief using primary sources and document unresolved uncertainty.”

The U.S. Department of Labor-sponsored ONET Content Model separates worker characteristics, worker requirements, experience requirements, occupational requirements, and occupation-specific information. It offers a useful public taxonomy for role analysis, though an organization must validate its own work context.

Build a role-to-capability map

For each role family, list critical tasks, the capability required, target proficiency, demand horizon, and consequence of a gap. Avoid copying a generic skill library without checking whether the task actually occurs.

Synthetic example:

| Task | Capability | Target evidence | Demand | Gap consequence | | --- | --- | --- | --- | --- | | Prepare decision briefs | Evidence navigation | Traceable brief with source-quality judgments | Current | Unsupported recommendation | | Review AI-assisted analysis | Verification and risk control | Error log and corrected output | Growing | Missed factual or operational risk | | Design assessment cases | Context framing | Approved scenario with constraints and success criteria | Current | Unrealistic or ambiguous assessment |

The target evidence describes what would demonstrate capability; it is not a score benchmark.

Use an evidence ladder

Classify each observation:

These levels answer different questions. Do not automatically convert course completion into verified workplace proficiency. Note recency, task similarity, evaluator consistency, and whether assistance was allowed.

• Self-report: person says they can perform the skill.
• Participation: person completed training or practice.
• Knowledge check: person explains concepts under defined conditions.
• Work sample: person produces a relevant artifact.
• Observed performance: person completes a realistic task with process evidence.

Worked skills finding

Suppose 40 employees report confidence in AI-assisted research, 32 completed a course, 18 submitted a work sample, and 11 demonstrated source verification in a realistic exercise. A weak analysis says, “80% of the workforce is proficient” by using course completion as the denominator and definition.

A responsible finding says: “Training participation is broad, while direct evidence of source-verification performance is available for only 18 people. Eleven demonstrated the defined behavior in the exercise. Evidence for the remaining population is missing, not negative.” The numbers are synthetic and should not be generalized.

Turn gaps into evidence plans

Separate three remedies: learning for a known knowledge gap, practice for an execution gap, and measurement for an evidence gap. If there is no relevant observation, gather evidence before prescribing training. Prioritize by task importance, expected demand, current evidence strength, and risk—not by the largest spreadsheet difference alone.

AI can map variant skill terms, surface missing role coverage, and draft development options. Human owners should confirm task relevance, fair access, accommodations, and interpretation. Do not use this analysis as an automated employment decision.

Verification checklist

Confirm role tasks with work owners; define observable behaviors; record evidence type and recency; preserve missingness; avoid false precision; review group coverage and accessibility; document evaluator limitations; and keep development decisions separate from employment determinations.

Sources and claim map

All roles, counts, and findings are synthetic. This resource does not establish a validated proficiency benchmark, assure training outcomes, or recommend hiring, promotion, or termination decisions.

• ONET Resource Center, The ONET Content Model: supports distinguishing multiple layers of occupational and worker information when structuring role analysis.
• ONET Resource Center, ONET database: provides the official downloadable occupational data resource and documentation for appropriately scoped reference use.

Sources: [1] [2]