Why the board meeting always ends with "how did you get this number?"
A People Analytics Lead presents a slide: "38% of the finance function is highly exposed to AI." The room nods for about four seconds. Then the CFO asks the only question that matters — how was that number produced? If the honest answer is "a vendor's model said so," the slide is dead on arrival. Numbers that can't be traced back to a visible method don't survive contact with a finance function, and they shouldn't.
This is the structural problem with most AI exposure claims circulating in HR right now: they arrive as a single output with no visible working. A transparent rubric fixes this by making the inputs, the weights, and the arithmetic inspectable — so the number can be defended, challenged, and recalculated by someone other than the person who built it. This article walks through what that rubric actually needs to contain.
The problem with a black-box exposure score
An exposure score that can't be decomposed is not a finding — it's an assertion. If a tool tells you a role is "72% exposed" and can't show you which tasks drove that number, or how much weight each task carried, the figure is unfalsifiable. You can't audit it, you can't override a single input you disagree with, and you certainly can't defend it to a board member who has read a headline about AI and jobs and wants a straight answer.
Workforce planning has a documented credibility gap already. 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 counting and that they struggle to demonstrate its return on investment. An opaque exposure score dropped into that environment doesn't close the credibility gap — it widens it, because it looks like one more black-box number nobody can defend under questioning.
The fix isn't a better black box. It's a rubric that shows its work at every step: which occupational data it started from, which dimensions it scored, how it weighted them, and how those component scores rolled up into the number on the slide.
The four dimensions: what a transparent rubric actually scores
A defensible rubric starts by refusing to score "the job" as a single blob. Jobs are bundles of tasks, and different tasks expose differently. A transparent framework scores each task against four independent dimensions, each on a 0–100 scale:
- Cognitive routine — how standardized and rule-based the reasoning in the task is.
- Physical routine — how repeatable and physically structured the task is.
- Social/judgment — how much the task depends on relationship management, negotiation, or contextual discretion.
- Creative — how much the task requires originating novel output rather than recombining existing patterns.
Scoring across four dimensions instead of one prevents the most common distortion in exposure analysis: treating "this role uses a computer" as a proxy for "this role is exposed." A financial analyst and a claims processor might both score high on cognitive routine, but they diverge sharply on social/judgment and creative — and a rubric that only measures one dimension erases that difference. Our four-dimension task exposure framework walks through how each dimension is defined and scored in detail, including the boundary cases that are hardest to call.
Weighting tasks by importance, not by guesswork
Scoring tasks is only half the method. A role is a bundle of tasks with different weights — some are core to the job, some are occasional. A transparent rubric borrows its task list and importance ratings from O*NET, the U.S. Department of Labor's public occupational database, rather than guessing at what a role "mostly" does.
O*NET covers 1,016 occupational titles (923 with full data-level detail, representing more than 55,000 individual jobs) and rates each task within an occupation on standardized importance scales, refreshed on a regular update cycle — quarterly, with a primary annual update typically landing in the third quarter. That importance rating becomes the weight applied to each task's four-dimension score before it rolls up into a role-level number.
This matters because it replaces analyst intuition with a documented, third-party-sourced weight. If someone challenges why a particular task moved the score more than another, the answer isn't "our model decided" — it's "O*NET rates that task as more central to the occupation, and here's the rating." Our importance-weighted task exposure scoring piece breaks down the weighting formula step by step.
From task score to role score: a worked example
Here is a simplified, illustrative version of the arithmetic — round numbers, not a real occupation, purely to show the method.
Suppose a role has three core tasks. Task A carries an O*NET importance weight of 4.5 (out of 5) and scores 80 on cognitive routine, 10 on physical routine, 20 on social/judgment, and 15 on creative. Task B carries a weight of 3.0 and scores lower on routine, higher on judgment. Task C carries a weight of 2.0 and scores highest on creative. Each task gets a composite exposure score from its four dimension scores, and each composite is then multiplied by its importance weight and divided by the sum of all weights — the same logic as a weighted average in any spreadsheet, just applied to four dimensions instead of one.
The output is a single role-level number, but critically, every input that produced it is still visible: the task list, the O*NET weight for each task, and the four dimension scores behind each composite. Anyone reviewing the number can trace it back to its parts and disagree with a specific input rather than the output as a whole. That traceability is the entire point of the exercise — see our AI exposure score meaning article for how to read and communicate the resulting number to non-technical stakeholders.
What a rubric can't tell you — and why that's the point
A transparent rubric will still only ever produce an exposure score — a measure of how much of a role's task content overlaps with what current AI systems can plausibly perform. It does not, and should not, predict who gets laid off, which roles will be automated, or which employees are "safe." Those are organizational decisions that depend on strategy, budget, retraining capacity, and dozens of factors no task-level rubric can see.
The right way to use an exposure score is as an input to a judgment process, not a verdict. A well-built rubric sorts roles into categories like Monitor, Review, and Redeployment Candidate — signals that tell a leadership team where to look closer, not conclusions that tell them what to do. Our task-based AI exposure methodology overview covers how those categories get assigned and what each one is meant to trigger internally.
Building or buying: what to look for in a documented rubric
Whether you build a rubric internally or evaluate a vendor's, the same checklist applies. Does it name its dimensions explicitly, or hide behind a single composite number? Does it disclose where its task list and importance weights come from? Can you see the arithmetic that turned a task score into a role score? Can you override a single input without recalculating the whole model by hand?
If the answer to any of those is no, the rubric is not transparent — it's just a more polished black box. WorkforceAnalysis was built around this checklist directly: the four-dimension scoring, the O*NET-sourced task weights, and the full audit trail from task to role to department are visible at every tier, not locked behind a support ticket. You can see how the tiers are structured on the pricing page, or start with the standalone AI Exposure Scorecard template if you want to run the rubric on a handful of roles before committing to a full department rollout.
This site incorporates information from ONET. Used under the CC BY 4.0 license. ONET is a trademark of USDOL/ETA.