How AI-enabled knowledge work differs across industries

Map tasks within sectors

Exposure to AI occurs at the task level, while most occupations contain a mix of activities. Sector analysis should identify which tasks may change, which remain dependent on human input and how the surrounding role coordinates evidence, judgment and delivery.

The ILO's 2025 global index evaluates generative AI exposure using task-level data across detailed occupations. It reports that transformation is more likely than complete replacement for most jobs because occupations combine tasks and continue to require human input.

For an organization, the practical unit is even more specific: a named workflow such as preparing a credit memo, analyzing service incidents or drafting a regulatory response. The same tool can have different value and risk in each workflow.

Sources: [1] [2]

Identify sector-specific evidence and constraints

A sector strategy should document authoritative sources, sensitive data, professional or legal obligations, material error types and required approvals for each workflow. These conditions determine what training, verification and governance must emphasize.

A financial analysis may depend on reproducible calculations and reporting periods. A healthcare-adjacent communication may require careful privacy and clinical boundaries. A public-sector recommendation may require policy traceability, accessibility and a reviewable decision record.

The OECD's research on changing skill demand shows that AI adoption can reorganize work and change demand for business, management, digital and social skills. Those changes should be examined empirically within the organization rather than inferred from a generic industry forecast.

Sources: [2] [1]

Measure capability within context

Cross-industry measurement needs a stable core and contextual layers. A common capability model can support comparison, while scenarios, evidence standards and interpretation rules should reflect the workflow and sector conditions that shape good performance.

A shared core might examine framing, evidence use, AI orchestration, verification, judgment and communication. The scenario should then supply realistic materials, constraints and consequences for the selected work. This preserves comparability without pretending every role performs the same task.

Early benchmarks should identify the population, assessment family and sample limitations. If evidence is insufficient for a sector comparison, report a directional or unavailable result instead of borrowing precision from a broader group.

• Start with tasks, not industry headlines.
• Map authoritative evidence and sensitive inputs.
• Define the consequences of material errors.
• Keep a common capability core with contextual scenarios.
• Label benchmark populations and limitations.

Sources: [1] [2]