A board question that a single headline number can't answer
Three weeks before a board meeting, a People Analytics Lead gets a message from the company's PE sponsor: "I read that Goldman says AI exposes 300 million jobs. What does that mean for us?" The number is real. It is also not built to answer that question.
This is the trap every HR strategy team eventually walks into. The Goldman Sachs AI jobs report is one of the most-cited pieces of AI-workforce research in circulation, and it belongs on the reading list of anyone doing this work. But it was built to estimate a global labor-market effect, not to tell a 900-person logistics company which roles in its finance department merit a closer look. This article walks through what Goldman actually published, how it compares to the other major macro studies your board has probably also heard about, and how to move from an economy-wide estimate to a defensible, company-specific answer.
What the Goldman Sachs AI jobs report actually measured
The figure most people remember from Goldman Sachs' 2023 research is that generative AI could expose the equivalent of 300 million full-time jobs globally to automation (Goldman Sachs, 2023). That number is a macroeconomic estimate tied to Goldman's broader thesis that generative AI could raise global GDP by roughly 7% over a decade. It is a top-down calculation built from occupational task data aggregated across economies, not a bottom-up audit of any single employer's org chart.
That distinction matters more than the headline number itself. A global estimate answers "how large is this phenomenon," which is useful for setting board-level context. It cannot answer "which of our 40 job families should we prioritize for review this quarter," because it was never built with your job families, your task mix, or your org structure as an input.
How Goldman's number compares to the rest of the macro research
Goldman is not the only major research shop that has published an economy-level estimate, and boards increasingly want to know how the numbers relate to each other. McKinsey has estimated that 60–70% of employees' work hours could theoretically be automated with current and emerging technology, up from roughly 50% in its pre-generative-AI estimates (McKinsey, 2023). The World Economic Forum's Future of Jobs Report 2025 frames the question differently again: it projects 170 million jobs created and 92 million displaced globally by 2030, a net gain of roughly 78 million jobs against 22% total labor-market churn (WEF, 2025).
These are three different research designs measuring three different things — theoretical automatable work-hour share, net job creation versus displacement, and equivalent full-time jobs exposed. They are not interchangeable, and a board slide that blends them without attribution is doing the reader a disservice. If your organization wants a fuller comparison of how the major studies frame the question, our breakdown of the McKinsey AI workforce report walks through that study's methodology in the same detail this article gives to Goldman's. We've also written a dedicated explainer on how many jobs AI will replace by 2030 that lines up Goldman, McKinsey, and WEF side by side with their original sourcing.
Which jobs the research says are most exposed
Beyond the headline totals, the macro research offers a consistent directional signal about where exposure concentrates. Brookings' 2024 analysis found that the highest-exposure sectors are STEM, business and financial operations, engineering, and law (Brookings, 2024), and separately estimated that roughly 12.9 million workers — about a third of those in the most-exposed occupations — are highly exposed to generative AI (Brookings, 2024). Anthropic's Economic Index found that AI usage was concentrated in specific occupational task types, with the share of occupations using Claude for at least a quarter of their tasks moving from 36% in early 2025 to 49% on a pooled basis (Anthropic, 2025).
The pattern across this research is clear: exposure is not evenly distributed, and it correlates more strongly with the composition of daily tasks than with job title or seniority. That is a genuinely useful macro insight. It is also, again, a population-level pattern rather than a company-specific finding. Our companion piece on which jobs are most exposed to AI goes deeper on the occupational categories the research flags most consistently, and our broader survey of AI job displacement statistics collects the full set of figures cited across this article in one reference table.
"300 million full-time jobs exposed globally" tells a board that the phenomenon is large. It does not tell them which department to review first.
Why a macro estimate can't do a board's homework
None of the research above — Goldman's, McKinsey's, WEF's, Brookings', or Anthropic's — was designed to produce a company-specific answer, and it is worth being explicit about why. Macro studies work from national or global labor statistics, occupational classifications, and survey data aggregated across industries. They are built to detect economy-wide shifts, not to score the specific task mix of your accounts-payable team versus a peer company's.
When a board or PE sponsor asks "which of our roles are exposed," they are asking a question that requires mapping your actual role list to standardized occupational task data, then scoring each role's task composition against a consistent rubric. That is a different exercise from reading a research summary, and it is the exercise this site's methodology was built around.
From economy-level research to a defensible company-specific answer
The bridge between macro research and a board-ready answer is occupational task data — specifically, the U.S. Department of Labor's ONET database, which covers 1,016 occupational titles across 923 data-level occupations, representing more than 55,000 individual jobs (O*NET Resource Center, 2019). ONET rates each occupation against roughly 277 descriptors and is updated on a regular cycle, with a primary annual update in the third quarter (U.S. DOL, 2025). Occupations in O*NET map to the 2018 Standard Occupational Classification system, which organizes 867 detailed occupations into broader groups used consistently across federal labor statistics (U.S. BLS).
Our methodology maps each of your organization's roles to its nearest ONET occupation, then scores that occupation's constituent tasks against four dimensions — cognitive routine, physical routine, social/judgment, and creative — each on a 0–100 scale, weighted by the task-importance data ONET already publishes for that occupation. A worked example makes the logic concrete: if a role's core tasks carry O*NET importance weights of 40% "process routine transactional data," 30% "resolve non-standard customer disputes," 20% "coordinate with cross-functional teams," and 10% "design new process documentation," each task gets scored independently on the four dimensions, and the role's composite exposure score is the importance-weighted average across all four. The output is never a verdict. It's a starting point for one of three categorizations — Monitor, Review, or Redeployment Candidate — that a People team then applies its own judgment to.
This site incorporates information from ONET. Used under the CC BY 4.0 license. ONET is a trademark of USDOL/ETA.
Turning research literacy into a board-ready position
Citing Goldman Sachs correctly — with its date, its scope, and its limits — signals to a board that HR is reading the research seriously rather than repeating a headline. But research literacy is the first step, not the last one. The next step is translating that literacy into your organization's own numbers: your role list, your task mix, your departments, scored against a transparent and repeatable rubric rather than a global estimate borrowed from someone else's model.
If you want a structured way to walk your leadership team through that translation before the next board cycle, our Workforce AI Readiness Assessment Guide lays out the questions to ask and the data to gather first. You can also see how the scoring tiers work on our pricing page.
Want the macro research explained like this — sourced, dated, and translated for HR — delivered as it's published? Subscribe to our newsletter and we'll send the next breakdown straight to your inbox.