When the board asks "which roles are exposed?"
A People Analytics Lead is three weeks out from a board meeting. The company's private equity owner has asked a version of the same question that is now standard in every portfolio review: which roles are exposed to AI, and what is the plan? The instinct is to reach for a headline statistic — a percentage from a research report, dropped into a slide. That number will not survive the first follow-up question, because it was never built to describe one company's role list. It describes the US labor market in aggregate.
The potential impact of AI on US occupations is real, well-documented, and uneven — but it is uneven at the task level, not the job-title level. Two roles with the same title can carry very different exposure, depending on what the people in those roles actually spend their time doing. This article walks through what the macro research actually shows, why it cannot be read as a company-specific verdict, and what a defensible, task-level view requires instead.
What "AI exposure" actually means at the occupation level
Occupation-level exposure research asks a narrower question than it sounds like it's asking. It does not ask "will this job disappear." It asks: of the tasks that make up this occupation, how many are structurally similar to what generative AI systems currently do well? Research from Goldman Sachs estimated the equivalent of 300 million full-time jobs globally carry tasks exposed to automation in this sense — a scale estimate, not a list of specific roles slated for elimination. McKinsey's 2023 analysis of generative AI's economic potential found that 60–70% of employees' time is spent on activities that could theoretically be automated with current and emerging technology, up from roughly 50% in prior estimates that excluded generative AI capabilities.
Anthropic's Economic Index, tracking real-world usage of its Claude models, found that by early 2025 a large share of tasks touching roughly 36% of occupations involved the model performing at least a quarter of the associated tasks, rising to 49% of occupations in pooled analysis. A separate 2026 Anthropic analysis of usage patterns found that 68% of observed usage occurred on tasks the model could fully complete, while only 3% occurred on tasks it could not meaningfully assist with at all. These are usage patterns, not verdicts about specific jobs — and they are exactly the kind of qualitative signal that should inform planning without being mistaken for a company-specific finding.
The O*NET architecture behind exposure research
Nearly all of this research — including WorkforceAnalysis's own methodology — is built on the same underlying data structure: the US Department of Labor's ONET database. ONET describes 1,016 occupational titles (923 of which have full data coverage) mapped across more than 55,000 individual jobs in the US economy, using roughly 277 descriptors per occupation that are updated on a rolling basis, with the primary annual update typically landing in the third quarter. Those descriptors include the specific tasks, skills, and work activities that make up each occupation — the level of detail that turns "financial analyst" from a job title into a list of discrete, individually assessable tasks.
The Bureau of Labor Statistics' 2018 Standard Occupational Classification system, which O*NET aligns to, organizes the US labor market into 867 detailed occupations, rolled up into 459 broad occupations, 98 minor groups, and 23 major groups. This hierarchy matters because it is the reason exposure research can be done at all in a structured, repeatable way — every occupation in the country has a common, task-level description that doesn't depend on how any one employer titles the role.
This site incorporates information from ONET. Used under the CC BY 4.0 license. ONET is a trademark of USDOL/ETA.
WorkforceAnalysis uses this same task architecture to build company-specific exposure scores, mapping each role in an organization to its closest O*NET occupation and then scoring the underlying tasks — not the job title — against four dimensions: cognitive routine, physical routine, social/judgment, and creative, each scored 0–100 and weighted by each task's documented importance to the occupation. For a deeper walkthrough of how that index is constructed, see our explainer on the AI occupational exposure index.
Why two roles with the same title can score differently
Here is a simplified, illustrative example — not a real company output, just round numbers to show the mechanism. Imagine two "financial analyst" roles at two different companies. Analyst A spends most documented task time on standardized reporting: pulling data, reconciling numbers against templates, producing recurring variance reports. Analyst B spends comparable time on the same job title but with tasks weighted toward client negotiation, judgment calls on non-standard deals, and cross-functional persuasion.
If cognitive-routine tasks scored 80 out of 100 for Analyst A's task mix, and social/judgment tasks scored only 25, the importance-weighted composite would land meaningfully higher on the exposure scale than Analyst B's mix — where social/judgment might score 75 and cognitive-routine only 30 — even though both hold the identical job title in the HR system. This is the entire reason job-title-level exposure claims fall apart under scrutiny, and it's why our guide on which jobs are most exposed to AI frames exposure as a task-composition question rather than a title lookup.
The macro research is consistent with this pattern at the sector level. Brookings' 2024 analysis identified STEM, business and finance, engineering, and legal occupations as the sectors with the highest generative AI exposure, estimating roughly 12.9 million workers — about a third of workers in those occupations — as highly exposed. McKinsey's Global Institute projected that by 2030, US employment mix will shift toward healthcare, STEM, and managerial roles, and away from customer service, office support, and food service roles. These are directional, sector-wide findings. They tell a People Analytics Lead where to look first — not what to conclude about any specific department.
From macro signal to structural change
The World Economic Forum's Future of Jobs Report 2025 offers the clearest picture of what happens after exposure is identified, and it complicates any narrative of simple job loss. The report projects 170 million jobs created and 92 million displaced globally by 2030 — a net gain of 78 million jobs, against a churn rate of 22% of total employment. It also found that 39% of workers' existing skills are expected to be transformed or become outdated by 2030, and that 63% of employers cite skill gaps as the single biggest barrier to workforce transformation.
Perhaps the most operationally useful figure in the report: of every 100 workers needing training as a result of these shifts, the WEF projects 59 will need it, but the outcomes split further — 29 are expected to be upskilled within their current role, 19 reskilled or redeployed internally, and 11 are unlikely to receive adequate training at all. That three-way split — upskill, redeploy, gap — maps almost exactly onto the vocabulary a task-level exposure tool should use: a role scoring high on routine dimensions is a candidate for Monitor or Review, prompting a closer look at training or task redesign; a role where automatable tasks dominate the composition and adjacent internal roles exist becomes a Redeployment Candidate for internal mobility planning. None of this labels a role "safe" or "automated" — it identifies where judgment and planning should be applied first. For a fuller picture of how these dynamics compound across a whole organization, see how AI is changing the future of work.
What this means for a company-specific view
Macro research answers "how big is this, generally." It cannot answer "which of our 340 roles need a closer look before next quarter's board meeting," because it was never built at the level of a specific company's role list, task mix, or org structure. That gap — between economy-wide exposure research and a defensible, department-by-department heatmap — is exactly what a task-level, O*NET-grounded assessment is for. Our explainer on AI exposure by occupation walks through how occupation-level data translates into that kind of company-specific view, and our pricing page outlines the tiers available for teams ready to move from macro research to a role-by-role assessment.
For teams that want a structured starting point before running a full assessment, the Workforce AI Readiness Assessment Guide walks through how to inventory roles and prepare task data ahead of scoring.
The research is clear that AI's impact on US occupations is uneven, substantial, and already measurable in usage data. What it is not — in any of the sources above — is a verdict on any single company's roles. That verdict, such as it is, belongs to the humans doing the planning, informed by task-level data rather than title-level assumptions.
Want a monthly digest of new occupational-exposure research as it publishes, translated into what it does and doesn't mean for workforce planning? Subscribe to our newsletter below.