The CEO Pay Ratio Doesn't Count the Workers Who Would Widen It
By Courtney Turrin · September 10, 2026 · 12 min read read
Category: For Curious People
A federal rule requires companies to disclose how far CEO pay has pulled away from everyone else's. The rule makes a company look better the moment it moves support staff to a vendor, an automated message, or an AI.
A federal rule requires companies to disclose how far CEO pay has pulled away from everyone else's. It turns out the rule makes a company look better the moment it moves its support staff from in-house to a vendor, an automated message, or an AI.
South Richmond-based tobacco company Universal Corp. did something unusual in its most recent proxy statement. Under a section titled "Our Ratio in Context," it explained why its CEO pay ratio number looked so bad.
Universal's CEO, Preston Wigner, made $3.17 million last year. The company's median employee, a seasonal tobacco worker in Mozambique, made about $3,200. That's a ratio of 995 to 1, the widest of any public company headquartered in the Richmond region. Universal's filing states that 95% of its 25,000 employees work outside the U.S., many in what the company calls "less developed countries and rural areas with lower wage geographies." For comparison, it provided a second number: measured against its median U.S. employee alone, at $52,291, the pay ratio drops to 61 to 1.
Universal didn't do anything unusual to produce that gap. It just employs its overseas workforce directly, which means every one of those workers legally counts toward this metric.
The Greater Richmond Region's CEOs Average 176 Times Their Workers' Pay
Richmond BizSense reported that CEOs at the 19 largest publicly traded companies headquartered in the region averaged $8.6 million in pay last year, against a median employee pay of $104,000, a ratio of 176 to 1. Henrico-based armored transport company Brink's shows the same pattern as Universal: a global workforce spread across more than 100 countries produces a $16,000 median pay and a 598-to-1 ratio, while its U.S.-only median of $80,762 produces a 123 to 1 pay ratio. Altria, the region's highest-paid CEO at $24.56 million, came in at 147 to 1, against a $167,000 median.
Search "CEO salary Richmond" and the results paint a different picture. Indeed's aggregator puts the average CEO salary in Richmond at $135,288, based on eight job postings, the highest-paying of them being a local nonprofit. That number is much lower than those reported in the BizSense article because Indeed's calculation is pricing the job title "CEO" across every small business, nonprofit, and solo operation that uses it, versus the 19 companies large enough to trigger a federal disclosure requirement. The real gap between CEO and worker pay is invisible to a casual search because it's concentrated in a small number of large companies.
Richmond's pattern is reflective of the national one, but there's a longer story about the pay ratio (Figure 1).
[datawidget key="pay-ratio-history-line"]
The AFL-CIO's 2026 Executive Paywatch report put the average S&P 500 CEO-to-worker pay ratio at 312 to 1 for 2025, up from 285 to 1 the year before. EPI's historical data puts the same ratio at 59 to 1 in 1989 and 21 to 1 in 1965. Whatever else has changed about the American economy since 1965, the distance between the top of a company and its median worker has widened by more than an order of magnitude.
Employee Status Decides What Counts Toward the Pay Ratio
None of Richmond's disclosures explain why the rule works this way. That answer is in the statute.
The pay ratio disclosure exists because of Section 953(b) of the Dodd-Frank Act, implemented as Item 402(u) of SEC Regulation S-K. It requires public companies to calculate the median pay of all employees, compare it to the CEO's pay, and disclose the ratio. The employee population covers every full-time, part-time, seasonal, and temporary employee of the company and its consolidated subsidiaries, anywhere in the world, which is why Universal's and Brink's overseas workforces count and drive their ratios up. The rule explicitly excludes independent contractors and leased workers who are employed by an unaffiliated third party.
That single distinction, employee versus non-employee, produces four different outcomes for what happens to the same support job (Figure 2).
[datawidget key="sec-four-way-split"]
Keep it in-house in the U.S., and it counts fully toward the pay ratio calculation. Move it to a subsidiary the company owns outright overseas, and it still counts, pulling the median down the way Universal's and Brink's do. Contract it to a vendor, and the worker disappears from the calculation even though that person is still answering the phone for the same company. Automate it, and there's no job left to count at all. Two of these four outcomes make a company's ratio look worse. Two make it look better. Only one of the four requires raising anyone's pay.
A 2020 analysis by researchers Paul Willman and Alexander Pepper at the London School of Economics found the same mechanism at work in the United Kingdom's version of this disclosure, describing it as a shift from "administered" to "outsourced" inequality: companies that could outsource their lowest-paid work reported dramatically better ratios than companies in industries where that wasn't practical.
How One Company's Median Pay Jumped Eightfold in a Year
Most companies that shift workforce toward the excluded outcomes don't get caught in the act. One did, because of a scheduling accident.
Concentrix, now one of the largest customer-experience outsourcing firms in the world, was a wholly owned business inside SYNNEX Corporation until it spun off as an independent company on December 1, 2020, one day after SYNNEX's fiscal year ended. Because the split happened after the books closed, SYNNEX was required to include Concentrix's workforce in that year's pay ratio filing. For fiscal 2020, with Concentrix's employees still counted, SYNNEX reported median employee compensation of approximately $6,934. For fiscal 2021, with Concentrix gone, the successor company, TD SYNNEX, reported median employee compensation of approximately $56,410 (Figure 3).
[datawidget key="synnex-median-employee-jump"]
The same call center employees were still ultimately working for the same company, but were no longer part of the pay ratio calculation. TD SYNNEX's 2021 filing also reflects a September 2021 merger with Tech Data, which changed the workforce independently of the Concentrix spin-off, so the ratio comparison isn't clean. The median pay figure is: it comes straight from each company's own filing, and it increased by roughly eightfold in twelve months without anyone getting a raise.
In its first full year as an independent filer, Concentrix disclosed a median employee annual compensation of $5,384 against CEO Chris Caldwell's $15.1 million, a ratio of roughly 2,802 to 1 (Figure 4). By its most recent filing, Concentrix reported more than 450,000 employees in 75 countries, with over 65% located in Asia-Pacific, Latin America, or Africa.
[datawidget key="concentrix-own-ratio"]
Every one of those workers answers calls for someone else's customers. None of them will ever appear in the pay ratio of the company whose customers they serve.
Americans Have Trained Their Own Replacements Before
The work now moving overseas has moved to lower-cost employees before, and the handoff wasn't always overseas.
In 2015, The New York Times reported that Disney had laid off roughly 250 IT workers and required many of them to train the H-1B visa holders replacing them as a condition of severance. The story wasn't offshoring; the replacement workers were brought into the country. But the shape of it, an experienced worker required to hand off their own job to whoever will do it cheaper, is the same thing now playing out in reverse, as jobs that were offshored a generation ago start disappearing on the other end.
India's technology trade association, Nasscom, projects the country's business process management sector will reach $59 billion in revenue for the fiscal year ending in March 2026, growing 6.1%, while headcount grows only 2.3% (Figure 5). The share of contact-center revenue coming from plain voice work is projected to fall from roughly half today to between 30% and 40%, as AI tools absorb routine calls.
[datawidget key="india-bpm-revenue-vs-headcount"]
India's largest IT firms added a net total of 17 employees across the first nine months of fiscal 2026, a sharp slowdown from prior years. The qualities that made this work easy to move overseas, repetition, predictability, and scale, are the same qualities that make it easy to automate, and the calls recorded over two decades of offshore support are now training the systems replacing that work. A junior coder in Bengaluru, laid off by email after being named star performer of the year, was told the decision came down to reduced need "because of AI." A quality analyst in Manila was let go after helping train the AI system that replaced his role.
This isn't a story about one country's jobs moving to another's. It's a story about the bottom rung of the ladder disappearing on both continents.
The Bureau of Labor Statistics Counts These Hours. Productivity Statistics Don't.
The customer service industry is shrinking by design. The Bureau of Labor Statistics projects customer service representative employment will decline 5% between 2024 and 2034, even as the industry adds hundreds of thousands of openings a year purely to replace people who leave. What that shrinkage produces on the customer's end of the phone is a transfer of labor nobody can bill for.
A February 2026 report by Groundwork Collaborative, co-authored by Stanford economist Neale Mahoney, found that time spent on the phone with customer service has risen 60% over the past two decades, citing the Bureau of Labor Statistics' American Time Use Survey (Figure 6). They estimate that this ends up costing Americans at least $165 billion per year in wasted time and lost money.
[datawidget key="consumer-phone-time-increase"]
The same report estimates that consumers lose $21.6 billion a year in time dealing with health insurance paperwork. The National Customer Rage Survey, run through Arizona State University, found that 74% of Americans reported a product or service problem in the past year, more than double the rate in 1976. Customers surveyed reported wasted time (56%), lost money (43%, an average of $1,261), and emotional distress (31%) as a result.
Economists have a term for this that predates AI entirely: shadow work, coined by the philosopher Ivan Illich in 1981 and popularized more recently by journalist Craig Lambert, describing the unpaid labor a company shifts onto a customer once a human employee is removed from the transaction. Pumping your own gas and bagging your own groceries were early versions. Untangling a denied insurance claim through an automated phone tree with no option to reach a person is a more recent one.
The hours aren't going uncounted. The Bureau of Labor Statistics measures them directly, in its time-use survey. But when the same agency calculates labor productivity, it measures output per paid hour, and unpaid customer time never enters that calculation. The economy registers as more efficient by that measure. The work didn't disappear; it moved onto the customer, unpaid and uncounted.
Some of the Cuts Are Already Being Reversed
Gartner predicted in February 2026 that by 2027, half of companies that attributed customer service headcount reductions to AI will rehire staff to do similar work, under different job titles (Figure 7). Gartner found that only about a fifth of service leaders had reduced agent staffing because of AI in the first place.
[datawidget key="gartner-rehire-different-titles"]
Klarna's reversal is the clearest documented case: after its CEO announced in 2024 that an AI agent was handling work equivalent to 700 employees, he told Bloomberg in 2025 that the company had gone too far, that focusing on cost had produced lower quality, and that Klarna was reopening hiring.
Looking closely at what the reversals restore, Klarna's rehiring model is remote, contract-based, and aimed at students and rural workers, not the jobs that existed before. HP announced in November 2025 that it will cut 4,000 to 6,000 positions by 2028, an initiative CEO Enrique Lores said would impact product development, internal operations, and customer support specifically, targeting $1 billion in annual savings. Salesforce cut 4,000 customer support roles, with its CEO saying plainly that he "needs less heads," a case Channel RVA covered in detail alongside evidence that most 2026 layoffs attributed to AI don't hold up as genuine automation stories once the underlying data is analyzed (Figure 8).
[datawidget key="reversal-comparison-table"]
A human may be reintroduced, but the original job quality is never recovered.
The Tax Code Has the Same Blind Spot
The pay ratio rule isn't the only policy that seems to incentivize automation or contracting out jobs.
Former Transportation Secretary Pete Buttigieg, in an August 2026 appearance on The Diary of a CEO, pointed to the difference between how the tax code treats income from wages and income from capital. He argued that since capital gains are taxed at a lower rate than wages, the tax code functions as a subsidy from the U.S. government that gives companies more reason to automate a job than they would otherwise have. Whatever one makes of Buttigieg as a prospective 2028 candidate, the mechanism he's describing is a documented feature of the tax code.
Neither the SEC's disclosure rule nor the capital gains rate was written with the goal of rewarding outsourcing or automation. Both have this effect anyway, structurally, without requiring any executive to have made a cynical decision. That distinction points to what would change the incentive: not a change of heart in any boardroom, but a change in what these numbers are designed to count. A pay ratio rule that included contracted labor would stop rewarding the cheapest way to shrink it. A productivity statistic that accounted for the hours it exports to customers would stop mistaking that export for economic progress.
Until then, the number gets better every time the job gets worse.
Disclosure: The author is co-owner of Xponent21, an agency that builds and sells AI products for marketing and content production.