The Complete Guide to Company-Specific AI Workforce Exposure

Why "which jobs will AI take" is the wrong question for HR strategy teams

Three weeks before the board meeting, a People Analytics Lead gets a one-line request from the PE-backed CEO: "Which of our roles are exposed to AI, and what are we doing about it?" There is no internal precedent for answering this. The last workforce plan was a headcount spreadsheet. The last AI conversation was about a chatbot pilot in customer support. Now the board wants a company-specific picture, and it wants it grounded enough to survive a CFO's follow-up question: "How did you get this number?"

That question — which is really a demand for a defensible, repeatable methodology rather than a headline statistic — is the right one, and it is different from "which jobs will AI take." The second framing invites a verdict. It asks a tool, a consultant, or an analyst to predict outcomes: who gets automated, who gets laid off, who survives. No credible methodology can make that prediction at the level of an individual role in an individual company, because outcomes depend on decisions leadership hasn't made yet — decisions about redeployment, reskilling, workflow redesign, and investment that a task-level exposure score can inform but not replace.

This guide is about the first framing: how to conduct an AI workforce assessment that produces a defensible, task-level, company-specific exposure picture — one built on a public occupational taxonomy, a transparent scoring rubric, and a workflow that ends in categories your leadership can act on, rather than a verdict they can't defend.

We'll walk through what a company-specific assessment measures, the O*NET foundation that makes it repeatable across companies and time, the four-dimension framework used to score exposure at the task level, a five-step process for running the assessment, and how to turn scores into board-ready action categories without ever crossing into outcome prediction.

What a company-specific AI workforce assessment actually measures

Most of the AI-and-jobs research your board has already seen operates at the economy or sector level. The McKinsey Global Institute estimates that 60–70% of employees' time is spent on activities that could theoretically be automated with technology available today, up from roughly 50% in its pre-generative-AI estimate. Goldman Sachs has estimated that generative AI could expose the equivalent of 300 million full-time jobs globally to some degree of automation. The World Economic Forum's Future of Jobs Report 2025 projects 170 million jobs created and 92 million displaced by 2030 — a net addition of 78 million, but a churn rate of 22% of today's total employment.

These figures are useful for framing the scale of change. They are not useful for telling your CFO which of your 340 marketing, finance, and operations roles need attention this fiscal year. A company-specific assessment closes that gap by working at the task level, inside your actual org structure, using your actual role list — not a sector average.

That means three things a good assessment must do that a research report cannot:

• Map your job titles to a standardized occupational taxonomy, so "Senior Revenue Analyst" and "Financial Planning Manager" are scored against the same task inventory a labor economist would recognize, not against an internal, ad hoc description.
• Score exposure at the task level, not the job-title level, because two roles with the same title in two different companies can have meaningfully different task mixes — and because within a single role, some tasks are highly exposed while others are not.
• Roll task scores up to department and company views that a board can read in one sitting, while preserving the underlying task detail an HR strategy team needs to design an actual response.

The rest of this guide walks through how each of those three requirements gets satisfied in practice, starting with the taxonomy.

The O*NET foundation: how occupational data makes this repeatable

The credibility of a company-specific exposure assessment rests on using an occupational classification that is public, government-maintained, and independent of any vendor's incentives. O*NET — the Occupational Information Network, maintained by the U.S. Department of Labor's Employment and Training Administration — is that foundation.

O*NET currently describes 1,016 occupational titles, of which 923 have full data-level detail, collectively representing more than 55,000 individual jobs across the U.S. economy. Each occupation is described using roughly 277 descriptors covering the tasks performed, the skills and abilities required, the knowledge domains involved, and the work context — and the database is updated on a regular cycle, with a primary annual update each third quarter, so the underlying task descriptions don't go stale the way a one-off consulting engagement's assumptions do. [Verify the specific database version — production release v30.3 as of this writing — against what's live at assessment time.]

O*NET occupations map onto the Standard Occupational Classification (SOC) system maintained by the U.S. Bureau of Labor Statistics, which organizes 867 detailed occupations into 459 broad occupations, 98 minor groups, and 23 major groups. That nesting matters operationally: it's what lets an assessment roll individual role scores up into department- or division-level views without losing the ability to drill back down to the task detail behind any single number.

Two things follow from building on ONET rather than a proprietary taxonomy. First, the mapping is auditable — any HR leader, board member, or outside consultant can look up the same occupation and see the same task list, because it's public. Second, the exposure logic can be explained without asking anyone to trust a black box: task importance and task content come from a federal database; only the exposure scoring rubric layered on top is the assessment provider's own methodology, and that rubric should be fully disclosed. Our O*NET explainer covers the taxonomy in more depth; the job-title mapping guide covers the practical mechanics of getting from a messy internal role list to clean ONET codes.

This site incorporates information from ONET. Used under the CC BY 4.0 license. ONET is a trademark of USDOL/ETA.

The four-dimension task exposure framework, with a worked example

Once a role is mapped to an O*NET occupation, the assessment scores each of that occupation's constituent tasks against four dimensions, each on a 0–100 scale:

• Cognitive routine — how much of the task is repeatable, rules-based information processing (data entry, standard reporting, routine calculation).
• Physical routine — how much of the task is repeatable physical or manual work.
• Social/judgment — how much of the task depends on interpersonal negotiation, persuasion, mentorship, or context-dependent judgment calls that resist codification.
• Creative — how much of the task requires original synthesis, novel problem framing, or generative ideation rather than pattern application.

Each task also carries an O*NET-derived importance weight — how central that task is to the occupation overall — so a task an incumbent performs constantly and that matters greatly to job performance counts for more than a task performed rarely.

Here is a worked example, using round numbers to illustrate the arithmetic rather than to assert a real occupational score. Consider a simplified two-task slice of a customer-facing role:

To get the occupation-level score for each dimension, the assessment computes an importance-weighted average across all of the occupation's tasks. For cognitive routine here: (80 × 70 + 60 × 40) / (80 + 60) = (5,600 + 2,400) / 140 ≈ 57. For social/judgment: (80 × 60 + 60 × 85) / 140 = (4,800 + 5,100) / 140 ≈ 71. The same weighting applies to physical routine and creative. The result is four dimension scores per occupation, not one blended number — which matters, because a role can be simultaneously high on cognitive-routine exposure and high on social/judgment, and collapsing those into a single figure would hide exactly the nuance a board needs to design a sensible response.

This is the mechanical core of what our four-dimension exposure framework explains in full, including how the four dimension scores are presented alongside each other (never averaged into one misleading composite) and how they feed the category logic described in section six below.

Two things are worth stating plainly here, because they are easy to get backward. First, a high cognitive-routine score does not mean a role "will be automated" — it means a meaningful share of the role's importance-weighted task content is the kind of repeatable information processing that automation tooling has historically targeted first. Second, a high social/judgment or creative score is not a guarantee of durability against any future technology; it is a description of the task content as it exists today, scored against a defined rubric. The score is an input a strategy team reasons with, not a verdict handed down.

How to conduct an AI workforce assessment: a five-step process

With the taxonomy and the scoring framework established, the assessment itself follows a five-step sequence.

Step 1 — Compile the role list. Pull the current headcount export from the HRIS, including job title, department, and (ideally) a short duty description for ambiguous titles. This is the input the entire assessment depends on, so time spent cleaning duplicate or vague titles here pays off downstream.

Step 2 — Map each role to an O*NET occupation. This is rarely a clean one-to-one match. Internal titles like "Growth Operations Manager" or "Client Success Partner" don't appear verbatim in O*NET; they get mapped to the closest occupational match based on actual duties, not title text. Our O*NET mapping guide walks through the decision rules for handling hybrid roles, newly created titles, and cases where a role spans two occupations.

Step 3 — Score task exposure. For each mapped occupation, pull the O*NET task list and importance weights, then apply the four-dimension rubric described above, tasked-weighted up to an occupation-level (and by extension, role-level) score on each of the four dimensions.

Step 4 — Roll scores up to department and company views. Individual role scores aggregate into department-level heatmaps using the SOC group structure, so a CFO or board member can see, at a glance, which functions carry the highest concentration of cognitive-routine exposure without needing to read 340 individual role scores.

Step 5 — Assign action categories, not verdicts. Each role's exposure profile is translated into one of three categories — Monitor, Review, or Redeployment Candidate — which is what a leadership team actually needs to decide what to do next. This step is covered in full in the next section, because it's the one most prone to being misused if it's skipped or shortcut.

Done properly, this sequence takes a fraction of the analyst-hours a manual Excel-and-O*NET exercise requires, precisely because steps 2 through 4 are mechanical once the rubric and mapping logic are built — which is the structural argument for using a purpose-built self-serve tool rather than rebuilding this pipeline by hand for every assessment cycle.

From exposure scores to action: Monitor, Review, Redeployment Candidate

This is the section where most AI-and-workforce content quietly breaks its own credibility, so it's worth being explicit about the constraint. A task-level exposure score is a measurement of task content against a defined rubric. It is not, and cannot be, a prediction of what will happen to a specific person's job. No methodology — ours or anyone else's — should claim otherwise, and any tool or report that tells you a role "will be automated" or is "safe" is making a claim its data cannot support.

What the score can responsibly do is sort roles into categories that map to different management actions:

• Monitor — the role's exposure profile is low-to-moderate and stable; no immediate action is indicated beyond normal workforce planning cadence.
• Review — the role shows meaningful concentration in one or more exposure dimensions and warrants a closer look: which specific tasks are driving the score, whether workflow redesign or tool adoption is already underway, and whether the role's task mix is likely to shift over the planning horizon.
• Redeployment Candidate — the role's task content is heavily concentrated in dimensions where task-level exposure is high, and the organization should proactively map internal redeployment paths — using the same O*NET occupational network to identify adjacent roles with meaningful skill overlap — rather than waiting for attrition or a reduction event to force the question.

That third category connects directly to why the WEF's skills data matters operationally: its Future of Jobs Report 2025 finds that of 100 workers needing training as a result of the changing skills landscape, 59 need it, but only 29 are upskilled into their existing role and 19 are reskilled and redeployed internally — leaving 11 without a clear internal pathway. The report also finds that 39% of workers' current skills are expected to be transformed or become outdated by 2030, and that 63% of employers cite skill gaps as the biggest barrier to workforce transformation. A Redeployment Candidate category that doesn't connect to an actual internal skills-adjacency map — the subject of our redeployment methodology explainer — is just a label. The point of the category is to trigger a specific next action: identify adjacent O*NET occupations with high skill overlap, and use that as the starting point for an internal mobility conversation.

It's worth noting that Brookings research finds the highest-exposure sectors tend to cluster in STEM, business and financial operations, engineering, and law — with roughly 12.9 million workers, about a third of employment in those occupations, classified as highly exposed. That is a sector-level finding, useful for context on where a board should expect Review and Redeployment Candidate concentrations to appear, but it is not a substitute for scoring your own roles — sector averages can mask wide variation between two companies in the same industry with very different task mixes.

Choosing a tool: build vs. consult vs. self-serve platform

Once an HR strategy team understands the methodology, the practical question becomes how to execute it. There are three broad paths, and each has a real structural tradeoff.

Manual consulting or in-house Excel-and-O*NET mapping. This is the incumbent approach most mid-market teams default to, because it requires no new software purchase. It can be genuinely customized to a company's specific role list. But it is not repeatable — every assessment cycle starts close to from scratch — and it is time-intensive, consuming analyst hours that scale with headcount. Independent consultants in this space typically bill in the range of $100–$350 per hour, with a median around $150–$200, according to ConsultFees' 2026 data; a full assessment engagement for a mid-market org, at that rate, is a meaningful multi-week commitment of billable time.

Enterprise workforce-analytics platforms. Tools like Orgvue (workforce digital twin and scenario modeling, including an AI automation-potential capability), Visier (predictive people analytics with pre-built workforce metrics), and Faethm — now part of SAP, positioned as an enterprise future-of-work scenario tool — are built for large organizations with correspondingly large analytics teams. None of these publish standard list pricing; all are sold on an enterprise-contract basis, which typically means a longer sales cycle and a budget tier built for organizations well above the mid-market headcount range. ChartHop, a mid-market people-analytics platform, does publish its packaging, but it has no O*NET integration, no AI-exposure scoring, and no redeployment recommendation feature — it solves adjacent org-chart and headcount-planning problems, not this one.

Self-serve, O*NET-grounded exposure platforms. This is the structural gap WorkforceAnalysis is built to fill: a tool that maps your role list to O*NET occupations, applies the four-dimension exposure rubric transparently, and produces department-level heatmaps and redeployment candidates — sized and priced for a mid-market HR team, without an enterprise sales cycle or a standing analytics function. It doesn't replace the judgment of your HR strategy leadership; it replaces the manual mapping and scoring labor that judgment currently has to wait on. Our comparison of AI workforce planning tools walks through the tradeoffs above in more detail if you're evaluating options before your next planning cycle.

Whichever path you choose, two things from the Gartner HR Priorities research are worth keeping in front of your leadership team as you scope the project: 86% of HR leaders report they have not implemented strategic workforce planning, and 66% say their organization's workforce planning is limited to headcount planning or struggles to demonstrate ROI. A task-level, board-ready exposure assessment is one of the more concrete ways to move past both of those gaps — provided it's built on a defensible methodology, not a headline number.

If you want to see how this maps against your own role list, the ROI calculator will estimate the analyst-hours a manual mapping exercise would take for your headcount, and you can book a demo or start directly from the pricing page to run a first department through the four-dimension framework yourself. For teams that want the underlying methodology as a standalone reference before they bring a tool into a board conversation, the Workforce AI Readiness Assessment Guide walks through the same five-step process in a format built for internal circulation ahead of a strategy session — including how to present the findings to your board without overstating what the scores can tell you.