A People Analytics Lead's spreadsheet has 40 job titles and one blank column labeled "AI exposure"
Three weeks before a board meeting, a private equity owner asks the same question every operating partner is now asking: which roles in this company are exposed to AI, and what is the plan? The People Analytics Lead opens a spreadsheet with every job title in the company and one empty column. The temptation is to fill it with a single number pulled from a headline statistic — a "60% of tasks could be automated" line from a research report, applied uniformly. That number was never about this company, and pasting it into a board deck without a method behind it is the fastest way to lose the room's confidence.
A defensible exposure score is not one number. It is four numbers, each tied to a specific kind of work, combined through a transparent and repeatable rule. This article lays out that structure: the four dimensions, how O*NET's own task-importance data feeds the weighting, and how the combined score becomes an input to a judgment call — never a verdict on any single role.
Why one exposure score per role is not enough
Most roles are not uniformly exposed to AI and automation. A financial analyst role might include highly automatable data-reconciliation tasks alongside a genuinely judgment-heavy task like presenting a variance explanation to a skeptical executive. Collapsing both into a single number erases the information a workforce strategy team actually needs — which parts of the role to redesign, and which to protect.
The task-based AI exposure methodology starts one level below the role: at the individual task, as ONET defines it. ONET — the U.S. Department of Labor's occupational database — currently documents 1,016 occupational titles across roughly 923 data-level occupations covering more than 55,000 job types, according to the O*NET Resource Center. Each occupation carries a structured task list, and each task carries an importance rating. That structure is the raw material for a rubric that can be applied consistently across an entire organization, not just to a handful of headline roles.
The four dimensions, defined
A transparent AI exposure rubric scores every task along four independent dimensions, each on a 0–100 scale:
- Cognitive routine. How standardized and rule-based is the thinking the task requires? A task that follows a repeatable procedure — reconciling two ledgers against a known chart of accounts — scores high. A task that requires synthesizing ambiguous, incomplete information into a novel judgment scores low.
- Physical routine. How standardized is the physical execution of the task, where physical execution is relevant at all? This dimension matters far more for operations, logistics, and field roles than for most office-based work, and it is scored independently rather than folded into the cognitive dimension.
- Social / judgment. How much does the task depend on reading a specific human relationship, negotiating, persuading, or exercising contextual judgment that has no clean rule set behind it? High scores here indicate work that stays anchored to a person, not a procedure.
- Creative. How much does the task require originating something new — a strategy, a design, a narrative — versus recombining existing patterns? This dimension is deliberately kept separate from cognitive routine because a task can be cognitively demanding and still highly patterned (tax-code application, for instance), while a much simpler-looking task can require genuine originality.
Each dimension is scored independently rather than averaged into a single "difficulty" number, because the four kinds of exposure do not move together. A task can be high on cognitive routine and low on social/judgment at the same time, and a redesign strategy for that task looks completely different than the strategy for a task that is high on physical routine and low on creative. Explaining cognitive routine tasks and how they interact with AI in isolation from the other three dimensions is what makes the scoring auditable — a reviewer can ask "why is this task an 80 on cognitive routine?" and get a specific, task-level answer, not a shrug.
Where the weighting comes from: O*NET task importance
Scoring each dimension is only half the method. The other half is deciding how much each task's score should count toward the role's overall exposure picture — because not every task in a job description carries equal weight in the actual work.
O*NET already solves this problem at the occupation level. Its database — currently on release v30.3, with roughly 277 descriptors updated on a regular cycle and a primary annual update each third quarter, per the U.S. Department of Labor — includes an importance rating for every task tied to every occupation. A task rated as critical to the job (say, "review loan applications for completeness and accuracy" for a loan officer) carries more weight in the role's overall profile than a task rated as minor or occasional.
Importance-weighted task exposure scoring takes each task's four dimension scores, multiplies them by that task's O*NET importance weight, and sums the results into a single role-level exposure profile — still expressed across the same four dimensions, not collapsed into one number.
A worked example
Consider a simplified role with three tasks, using round numbers purely to illustrate the arithmetic:
| Task | Importance weight | Cognitive routine | Physical routine | Social/judgment | Creative |
|---|---|---|---|---|---|
| Reconcile monthly transaction reports | 0.40 | 80 | 10 | 15 | 10 |
| Prepare variance summary for review | 0.35 | 55 | 5 | 40 | 30 |
| Present findings to department head | 0.25 | 20 | 5 | 85 | 25 |
The weighted cognitive-routine score for this role is (0.40 × 80) + (0.35 × 55) + (0.25 × 20) = 32 + 19.25 + 5 = 56.25. Run the same weighted sum down each column and the role ends up with a full four-dimension profile — moderate cognitive routine, low physical routine, moderate social/judgment, low-to-moderate creative — rather than a single blended figure that would hide the fact that one task in this role is genuinely relationship-dependent while another is close to fully standardized.
This is exactly the kind of calculation a CFO or a board member is entitled to ask about, and exactly the kind of calculation a single "AI exposure: 62%" headline cannot answer.
From dimension scores to a category — never a verdict
Once a role has a four-dimension profile, the next step is translating it into something an HR strategy team can act on without over-claiming what the number means. That translation uses three categories: Monitor, Review, and Redeployment Candidate.
A role or task landing in Monitor shows a profile that does not currently warrant redesign — the mix of dimensions suggests the work is not concentrated in high-automation-potential territory today. Review flags a profile worth a structured conversation: perhaps cognitive routine is high but social/judgment is also high, meaning a partial redesign — automating the routine slice, protecting the relationship slice — is worth exploring. Redeployment Candidate flags tasks or task clusters where the profile suggests a genuine redesign conversation belongs on the roadmap, paired with the O*NET occupational category the role maps to so the redeployment options offered are grounded in real occupational adjacency, not guesswork.
None of these three categories is a prediction of what will happen to a role or the person in it. They are inputs to a human planning conversation — the same way a credit score is an input to a lending decision, not the decision itself. A role scored as a Redeployment Candidate is not "being automated" and the person in it is not "being replaced." The category describes where a structured redesign conversation is most likely to be productive, and the four-dimension breakdown underneath it tells the team exactly which part of the job to focus that conversation on.
Why this replaces a spreadsheet exercise, not a judgment call
The value of a transparent four-dimension rubric is not that it produces a scarier or more precise-looking number. It is that it produces a number anyone in the room can trace back to its source: which O*NET tasks fed it, what importance weight each task carried, and what evidence justified each of the four dimension scores. That traceability is what turns an exposure score from a talking point into something a board, a PE owner, or a skeptical department head can actually interrogate — and trust.
This site incorporates information from ONET. Used under the CC BY 4.0 license. ONET is a trademark of USDOL/ETA.
For teams that want to apply this exact rubric to their own role list without building the scoring logic from scratch, the AI Exposure Scorecard template walks through the same four dimensions and the same importance-weighting math role by role.