Why "which roles are AI-exposed" is the wrong question to open with
Three weeks before the board meeting, the request usually arrives in a single line from the audit or strategy committee: give us a picture of where AI changes our workforce, and what we're doing about it. It sounds like a data question. It is actually a governance question, and the distinction matters for how you build the deck.
A board that asks about AI workforce exposure is not asking for a list of jobs to eliminate. It is asking whether management has a repeatable, defensible process for understanding a structural shift before it becomes a crisis. If you walk in with a spreadsheet of role titles and gut-feel risk ratings, you will survive the first slide and lose the room on the second question. If you walk in with a methodology — one you can explain, defend, and repeat next quarter — you change the conversation from "are we exposed?" to "how are we managing this?"
This article lays out a framework for that second kind of presentation: what a defensible, company-specific exposure picture looks like, how to build it on a transparent rubric rather than a black-box score, and how to answer the follow-up questions a CFO or audit committee will actually ask. For a fuller walkthrough of the underlying exposure methodology, see our company-specific AI workforce exposure guide.
What boards actually want: a company-specific, defensible exposure picture
Most of what board members have read about AI and jobs is macro research — useful for context, useless for governance. The Goldman Sachs estimate that generative AI could affect the equivalent of 300 million full-time jobs globally, or McKinsey's finding that 60–70% of work hours could theoretically be automated with current technology, are directionally important. They tell a board that the category is real. They tell a board nothing about whether the underwriting team or the customer support group at this company is affected, or how.
That gap — between "AI is transforming the economy" and "here is what it means for our 40 roles in claims processing" — is exactly what a board wants management to close. And it wants that answer built on a method it can interrogate, not a consultant's confidence or a vendor's proprietary black box.
Two structural facts make this harder than it should be. First, Gartner's 2024 HR Priorities research found that 86% of HR leaders have not implemented strategic workforce planning at all, and a related Gartner survey found 66% of organizations say their workforce planning is limited to headcount planning or struggle to demonstrate its ROI. Most companies bringing an AI workforce story to the board are building the underlying planning discipline at the same time. Second, the World Economic Forum's Future of Jobs Report 2025 projects that 39% of workers' core skills will be transformed or become outdated by 2030, and that 63% of employers cite skill gaps as their biggest barrier to workforce transformation. The board is not just asking "who is exposed" — it is implicitly asking whether the organization has a plan for the skill shift, not only the task shift.
None of this justifies presenting speculative numbers with false precision. It justifies presenting a rubric.
The four-dimension exposure rubric your board will ask you to defend
A defensible exposure picture starts from occupational data, not job titles. The U.S. Department of Labor's ONET database — 1,016 occupational titles covering more than 55,000 real-world job types, built from roughly 277 descriptors per occupation and updated on a regular cycle — is the standard public reference for what a job actually involves at the task level. This site incorporates information from ONET. Used under the CC BY 4.0 license. ONET is a trademark of USDOL/ETA.*
Mapping a company's roles to their nearest O*NET occupations gives you a task list for each role, along with the U.S. Bureau of Labor Statistics' Standard Occupational Classification structure — 867 detailed occupations organized into 459 broad occupations, 98 minor groups, and 23 major groups — as the taxonomic backbone. That is the data layer. The judgment layer is the rubric applied on top of it.
WorkforceAnalysis scores every task in a role against four dimensions, each on a 0–100 scale:
- Cognitive routine — how repeatable and rule-based the reasoning involved is.
- Physical routine — how repeatable and rule-based the physical actions involved are.
- Social/judgment — how much the task depends on negotiation, persuasion, mentoring, or contextual human judgment.
- Creative — how much the task depends on original synthesis, novel problem framing, or aesthetic and strategic judgment.
Each task within a role also carries an O*NET importance weight — a measure of how central that task is to the occupation as a whole. The role-level exposure score is the importance-weighted combination of task scores across all four dimensions, not a single "AI risk" number pulled from a vendor's undisclosed model.
This is the sentence to put in front of your board, verbatim if useful: exposure is an analytical input to human judgment, not a verdict on any role. A score does not say a job will be automated, and it does not say a job is safe. It says how much of the work, as currently structured, resembles the kind of task that current AI systems perform well — and it hands that finding to your leadership team to interpret, not to an algorithm to decide.
A worked example: scoring one role end to end
Boards trust methodology they can follow with a pencil. Here is a simplified, illustrative example — round numbers, not a real company's output — showing how the math works.
Take a mid-level claims-processing role. Say the role maps to an occupation with four representative tasks, weighted by O*NET importance data as follows: reviewing claim documentation against policy rules (importance weight 0.35), communicating decisions to policyholders (weight 0.25), investigating ambiguous or disputed claims (weight 0.25), and mentoring junior adjusters (weight 0.15).
Now score each task on the four dimensions (illustrative values):
| Task | Cognitive routine | Physical routine | Social/judgment | Creative |
|---|---|---|---|---|
| Review documentation (0.35) | 85 | 5 | 15 | 10 |
| Communicate decisions (0.25) | 40 | 5 | 75 | 20 |
| Investigate disputes (0.25) | 55 | 5 | 60 | 45 |
| Mentor juniors (0.15) | 20 | 5 | 80 | 30 |
Importance-weighting the cognitive-routine column: (0.35 × 85) + (0.25 × 40) + (0.25 × 55) + (0.15 × 20) = 29.75 + 10 + 13.75 + 3 = 56.5. Repeat for each dimension and you get a four-number profile for the role, not a single score. That profile is what belongs in front of your board — a role that is high on cognitive-routine exposure but also carries meaningful social/judgment weight looks structurally different from a role that is high on cognitive-routine exposure and low everywhere else, even if a lazier single-number model would show them as equivalent.
From there, roles land in one of three categories — never "safe" or "will be automated," always: Monitor (low current exposure, worth revisiting on the standard review cycle), Review (moderate-to-high exposure on one or more dimensions, warranting a closer look at task redesign or training investment), or Redeployment Candidate (exposure profile plus organizational context suggests the role's current task mix is a strong internal-mobility conversation, not an elimination decision). That vocabulary is deliberate. It keeps the finding in the register of workforce planning, where it belongs, and out of the register of a layoff list, where it does not.
Anticipating the CFO's follow-up questions
Every board presentation on this topic gets the same handful of questions. Prepare answers before you walk in.
"How did you get this number?" Answer with the rubric, not a vendor's confidence. Walk through the four dimensions, the O*NET importance weighting, and show one worked example the way we did above. A board member who sees the arithmetic trusts the conclusion more than one who is asked to take it on faith.
"Is this saying we're going to cut headcount?" No. Say so directly. The finding is an exposure profile — an input to workforce planning decisions your leadership team makes with full context the rubric doesn't have: succession plans, pending reorganizations, retention risk, local labor market conditions. Reinforce that "Redeployment Candidate" is a flag for a conversation about internal mobility and training, not a decision that has already been made.
"What's the cost of not doing this?" Here, cite what's actually measured, and be precise about what each figure covers. SHRM's research puts the fully loaded cost of replacing an employee at 50% to 200% of their annual salary, and SHRM's benchmarking work found an average cost-per-hire of $4,129. Separately, the Josh Bersin Company's research on internal mobility found that external hiring costs roughly 3 to 5 times more than filling a role internally. These are independently sourced hiring-and-turnover benchmarks — cite them as exactly that, not as a projection of what your company's AI exposure work will save, which is a distinct and unverified claim you should not make in the room.
"What's everyone else doing?" Context, not competitive benchmarking against a specific rival: PwC's 27th Global CEO Survey found 25% of CEOs expected to reduce headcount by 5% or more due to generative AI, alongside 39% who expected to increase headcount by 5% or more — a reminder that the aggregate picture is not uniformly a reduction story. PwC's 28th Global CEO Survey, a year later, found 13% of CEOs reported an actual headcount reduction attributable to generative AI (16% in the most-affected sectors), against 17% who reported an increase. Brookings' analysis found the highest-exposure sectors cluster in STEM, business and finance, engineering, and law — useful for framing where your board should expect the sharpest task-level shifts, not as a claim about your specific roles.
Exposure is an analytical input to human judgment — never a verdict on a role, a team, or a person.
Keep that line as your anchor through every follow-up. It is the difference between a board that trusts the finding and a board that starts worrying about what you're not telling them.
Structuring the board deck: from heatmap to action plan
A board deck on this topic works best in four sections, in this order.
1. The methodology, in one slide. Four dimensions, O*NET task-importance weighting, the three-category output vocabulary. This is the slide that earns trust for everything after it — do not bury it in an appendix.
2. The department-level heatmap. Aggregate role-level scores up to department view. Boards think in departments and cost centers, not job titles. Show the distribution — how many roles land in Monitor, Review, and Redeployment Candidate — rather than a single company-wide average, which flattens exactly the nuance the rubric was built to preserve.
3. What this does and doesn't tell us. One slide, stated plainly: this is exposure, not a forecast of who leaves. It's a structured way to prioritize where training, task redesign, and internal-mobility conversations should happen first. It does not replace succession planning, retention strategy, or legal review of any workforce action — and any of those next steps should be verified with the relevant internal or external counsel before execution.
4. The action plan and next review cycle. Tie each Review and Redeployment Candidate finding to a concrete next step — a training investment, a task-redesign pilot, a mobility conversation — and commit to a re-scoring cadence. O*NET's underlying data updates regularly, with a primary annual update cycle; your internal review should run on a comparable cadence so the board sees this as a standing practice, not a one-time exercise.
What exposure findings should — and should not — commit you to
The single biggest risk in this presentation is not an underbaked methodology — it's overclaiming what the methodology proves. Resist the temptation to promise the board a return-on-investment figure for the exposure work itself, a headcount-reduction target, or a timeline for "AI replacing" any function. None of those are things a task-level exposure rubric can responsibly produce, and a board that catches you overstating it once will discount everything else you present.
What you can commit to, credibly: a repeatable process for identifying where task composition is shifting, a shared vocabulary the leadership team uses consistently (Monitor, Review, Redeployment Candidate), and a standing cadence for revisiting the picture as both the underlying occupational data and your own organization change. The WEF's Future of Jobs Report 2025 estimates that of every 100 workers needing new skills by 2030, 59 will need training — 29 upskilled in their current role, 19 reskilled or redeployed internally, and only 11 unlikely to receive the training they need. That's a useful frame for the board: the work ahead is overwhelmingly about training and redeployment, not elimination, and a credible internal process should reflect that ratio in how resources get allocated.
Building this deck from scratch — the methodology slide, the heatmap template, the "what this doesn't tell us" language, the action-plan structure — takes real time to get right, and getting the framing wrong in front of a board is expensive to walk back. Our Board Deck & Executive-Summary Template Pack gives you a starting structure built around exactly this framework, so you can spend your prep time on your company's findings instead of the slide architecture. You can also see current plan options on our pricing page, browse the rest of our template library, or join the waitlist if you're planning a board presentation for next quarter and want early access to new material as it ships. For the fuller methodology this deck structure is built on, our blog covers the exposure framework in more depth.
The board doesn't need you to have solved AI workforce risk. It needs to see that you have a defensible way of looking at it, and a plan for looking again.