Three Million Missing Jobs
The debate over AI and employment usually focuses on whether the number of jobs in a certain group of occupations or industries has been declining. To gauge the impact of AI on jobs, the real question isn’t whether employment is rising or falling. It’s whether employment is where we’d expect it to be if generative AI had never arrived.
That requires a counterfactual.
Take the five industries at the heart of the white-collar U.S. economy: Finance, Insurance, Information, Professional Services, and Business Services—what I call FIIPB. In practice, that covers much of the office and professional economy: banking, insurance, tech and media, consulting, legal and accounting, and administrative support. Together they account for more than 40% of GDP and have long served as the main entry point for college-educated workers beginning their careers.
From 2015 to 2019, FIIPB employment grew at a remarkably steady 1.9% per year. That steadiness across five years, two Fed tightening cycles, and multiple distinct sectors suggests a structural growth rate, not a cyclical sugar high. Extend that trend through 2026 and you get a clean benchmark for where white-collar employment would likely be today if the old trajectory had continued.
Instead, FIIPB employment is running roughly 3 million jobs below that path—about 8% short of trend. This is an approximation, not a forecast. But even a generous haircut leaves you with a 2 million job deficit—and a deficit that’s still growing.
And the main mechanism isn’t mass layoffs.
These 3 million missing jobs are mostly roles that were never opened, attrition that was never backfilled, junior positions that quietly disappeared from hiring plans. A manager decides not to hire this quarter, then makes the same decision again next quarter, and again after that. Over time those choices add up. None of it shows up in layoff statistics. It’s visible only in the distance between where employment is and where it should be.
FIIPB aggregate employment vs. pre-pandemic trend projection. Source: BLS CES.
That gap is hard to dismiss as lingering pandemic distortion. These sectors recover. By late 2022, FIIPB employment essentially reached the pre-pandemic trend line.
And then the pattern changed.
Employment peaked, flattened, and began drifting further below trend. By early 2026 the divergence wasn’t narrowing. It was widening.
Gap between actual and counterfactual employment (thousands) by FIIPB sector, Jan 2020–Feb 2026.
The shortfall isn’t confined to one industry. Professional and Business Services accounts for roughly 2 million of the 3 million missing jobs. Financial Activities is short about 735,000. Information is short about 246,000. The popular narrative—that this is a tech overhiring correction from 2021–2022—mistakes a subplot for the whole story. Information, the sector closest to “tech,” accounts for less than a tenth of the total gap. The sectors driving the shortfall are banking, insurance, consulting, legal, and administrative support. They didn’t overhire. They simply stopped hiring.
Information: the sector closest to “tech.” Its 246K gap is real but accounts for less than a tenth of the total shortfall.
Financial Activities: banking and insurance, short 735K jobs against trend.
Professional & Business Services: the largest FIIPB sector and two-thirds of the total gap.
Gap as percent of counterfactual. All three FIIPB sectors converge to roughly −8% by early 2026.
Over this same period, output in these sectors continued growing while employment stalled, producing a sustained productivity acceleration that doesn’t show up at the same magnitude elsewhere in the economy.
This doesn’t prove AI is the sole force at work. The economy experienced other developments. But if no-structural forces were responsible, you’d expect a gap that narrows as conditions stabilize. Instead, the gap is widening, and it’s widening with remarkable consistency across very different white-collar industries. The pattern is highly consistent with a broad-based productivity shock concentrated in knowledge-work sectors—exactly the sectors where generative AI has the most obvious applications.
Three million FIIPB jobs didn’t disappear. They just never arrived.
Methodology Notes
Data: BLS Current Employment Statistics (CES), seasonally adjusted, national series. Sectors: CES5000000001 (Information), CES5500000001 (Financial Activities), CES6000000001 (Professional & Business Services). Counterfactual: Log-linear trend fit on monthly data, January 2015–December 2019, projected forward from December 2019 levels. Gap = actual − counterfactual.