# The AI Payoff Hasn't Reached Most Companies. Some Raises Slowed Anyway.

*An Apollo economist says big companies outside tech show no profit gain from AI yet. His firm's July study found slower pay growth in the jobs AI touches most.*

By Jay Cadmus | Published October 26, 2026
Canonical: https://theainextwave.com/editions/2026-10-26/

## The Short Version

Picture a customer service rep at a regional insurance company, walking into her annual review this week. Her team got an AI assistant in the spring. Calls wrap up faster, and her manager says so. Her raise still comes in smaller than she'd hoped.

Two pieces of research from the same Wall Street firm speak to that meeting. On Saturday, Torsten Slok, chief economist at the investment firm Apollo Global Management, wrote that AI has produced no visible payoff for large U.S. companies outside the tech giants. Revenue per worker and profit margins at those companies have gone flat. In July, Slok and a colleague published a study estimating that pay in the jobs most exposed to AI, customer service among them, grew about 6.7 percentage points slower than in other jobs after 2023, while employment held steady.

So the gains are hard to find on company books, and pay in a handful of exposed jobs is already lagging. Those two findings sit awkwardly together. This edition walks through what each one shows, where each is thin, and what a worker can ask for before the next review.

## What Slok Sees Outside Big Tech

Slok's note ran Saturday in Apollo's daily newsletter, and by Sunday Bloomberg and TechCrunch had picked it up. It splits the S&P 500 in two. The "Magnificent 7" are Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla. The other 493 are the banks, retailers, insurers, drugmakers and manufacturers that make up the rest of the index.

In his charts, revenue per worker and profit margins are climbing for the seven giants. For the other 493, both lines are flat, and he sees no sign yet that AI is lifting productivity growth. "AI may be a great calculator, but in business, intelligence is mostly judgment," he wrote.

His explanation is about cost and trust. Mainstream companies, he wrote, are starting to treat token spending, the metered fees they pay each time they use an AI model, as "a cost center that weighs on margins." Bloomberg reported his view that these companies have little need for the newest models and won't pay for them. He also argues that AI can't carry accountability, culture or ethical judgment, which shape a company's reputation and, over time, its sales. His bottom line: "the AI story will be decided in the S&P 493, not the Mag 7."

Bloomberg placed the note against a nervous start to earnings season. Meta has said it will raise spending on AI, and its shares fell. Bloomberg also cited a Gartner estimate that more than $3.5 trillion in AI infrastructure spending could be written down by 2030 if AI revenue fails to justify it.

This is one economist reading broad charts. Margins and revenue per worker move with prices, interest rates, the mix of industries and the timing of adoption. Flat lines leave room for plenty of successful projects inside individual firms. Any gains are still too small to see from that altitude.

## Where the Pressure Shows Up First

Apollo's July 30 white paper, by Slok and Sania Edlich, looked at pay. The authors used government wage data from 2015 through 2025 and a measure of AI exposure from the Anthropic Economic Index, which tracks the work tasks people actually hand to Anthropic's Claude assistant. A job counted as highly exposed if AI showed up in at least half of its weighted tasks. Eleven of the 321 occupations they could match cleared that bar, covering about 5.8 million workers, roughly 3.7 percent of the labor force.

The list reads like a tour of office work. It includes customer service representatives, data entry keyers, medical records specialists, medical transcriptionists, market research analysts, wholesale sales reps, financial analysts, software testers and computer programmers.

After 2023, the authors estimate, inflation-adjusted pay in those jobs grew about 6.7 percentage points slower than in low-exposure jobs. They found no statistically significant change in employment. The gap was widest in the lowest-paid group of jobs, at about 10.7 percent, with no measurable effect in the best-paid group. Their conservative lower-bound estimate of lost wages is $28 billion a year.

The authors read this as employers capturing AI's gains by holding down pay while keeping staff. That's their interpretation, and the paper lists its own limits. The exposure measure comes from a single AI company's usage logs, because Anthropic is the only provider that has released such data for public research. Only 321 of roughly 800 government job categories could be matched. And the years after 2023 overlap with the post-pandemic reshuffling of the job market, which the authors say may account for part of the gap.

## Two Findings That Pull Against Each Other

Put the two notes side by side and a puzzle appears. If employers in exposed jobs are pocketing AI's gains through smaller raises, the savings should land somewhere. Slok's own October charts say they haven't reached the margins of big companies outside tech.

There are a few ways to square it. The wage study covers jobs across the whole economy, while the margin charts cover 493 large companies, so the two groups only partly overlap. Time saved could be going into lower prices or into absorbing other costs before it ever reaches profit. Slower pay in these jobs could also have causes besides AI, which the authors concede. No source I opened ties the same companies, workers and AI budgets together tightly enough to settle it.

Other evidence leans more hopeful on the company side. A working paper published in March through the National Bureau of Economic Research surveyed nearly 750 executives. They reported labor productivity gains that vary by industry, with the largest in high-skill services and finance, and they expected those gains to strengthen this year. The authors found executives perceive larger gains than the official numbers capture, likely because the revenue arrives later. They saw little sign of near-term job losses overall, along with a decline in routine clerical roles.

The official data are murky too. Reuters reported Monday that output per hour has posted some strong recent jumps, but the figures swing a lot, and some officials credit pandemic-era habits like mobile ordering. AI's effect on jobs, prices and output, Reuters wrote, remains to be seen.

My read is that the payoff and the pay question run on different clocks. Companies are still paying to learn the tools, so the books show the cost first. Workers in a few exposed jobs may feel the squeeze sooner, in the size of a raise, long before anyone can prove where the saved time went.

## Dana's Annual Review

Take Dana, an illustration, who answers claims calls for a regional insurer. Last spring her team got an AI assistant that drafts replies and pulls up policy details. Her manager now points to faster calls as the reason the team can absorb more volume. Her raise came in below what she'd expected.

What Dana needs first is a baseline. She can ask what her team's numbers looked like before the tool arrived: calls per hour, how many customers called back about the same problem, and how often supervisors had to fix an answer. Without those, nobody can say whether the assistant helped or simply sped up the same mistakes.

Then she can ask how results are counted. A call closed fast and reopened two days later saved no one anything. The count that matters is accepted work: claims resolved without a callback or a correction. She should also name the added work, because checking the assistant's suggested answer on a complicated claim takes real minutes that rarely show up in a productivity chart.

A small, bounded way to test this is a single quarter. Dana and her manager agree on the baseline, track accepted calls and correction time, and revisit pay when the quarter ends. It can fail if the tool's answers are wrong and she carries the blame, or if saved time just becomes more calls at the same pay. Her manager and the company's compensation team hold the authority, so she asks for the criteria in writing.

## Opportunity Radar

Small and midsize employers that buy AI seats may have no baseline for what comes back, and their best staff can now point to research saying exposed workers' pay is lagging. An HR consultant or fractional finance person could offer a flat-fee "AI dividend review": set a baseline for one team, measure accepted output and correction time for a quarter, and propose how to share any real gain through pay, hours or training. The pitch is to owners who want to keep good people. Demand is a hypothesis, so test it by offering the review to five local firms you already know and see whether two sign up. Walk away if the measured gains come back near zero, because then there's nothing to share.

## What You Can Do With This

### If you work in one of the jobs on Apollo's list

Customer service, data entry, medical records, transcription, market research and similar roles are where the study found the biggest pay gap. Keep your own record of what the AI tools let you do that you couldn't before, and bring it to your review as evidence of added value.

### If you're weighing a raise, review or job offer this fall

Ask how the company measures productivity since AI tools arrived, and whether time saved has changed anyone's pay. A clear answer tells you a lot about whether the gains will be shared.

### If you manage a team using AI tools

Set a before-and-after baseline and count accepted work, along with the time people spend checking the tool. If you're asking more of people because of AI, be ready to show what they get back.

### If you own a small business

Track what you spend on AI seats against what you can actually measure coming back. If the answer is unclear after a quarter, that's useful information too.

## The Bigger Picture

In the national numbers, AI so far shows up mainly as a building boom: data centers, chips and the debt to pay for them. The tech giants show the gains. The rest of corporate America shows the bill, and in a few office jobs the early pressure appears to be landing on pay. My read is that much of the next year will play out in ordinary review meetings, where someone asks what the tool saved and who got the difference. Workers who walk in with a baseline will have a better chance of getting a share.

## References

[Apollo, The Daily Spark: No Signs of AI Payoff Outside the Tech Sector (Oct. 24, 2026)](https://www.apollo.com/wealth/insights-news/insights/daily-spark/no-signs-of-ai-payoff-outside-the-tech-sector)
[Bloomberg via Yahoo Finance: Apollo's Slok Says AI Has Delivered Zero Payoff Outside of Tech (Oct. 25, 2026)](https://finance.yahoo.com/technology/ai/articles/apollo-slok-says-ai-delivered-092836159.html)
[TechCrunch: AI delivers 'zero financial payoff' to most companies, Apollo economist says (Oct. 25, 2026)](https://techcrunch.com/2026/10/25/ai-delivers-zero-financial-payoff-to-most-companies-according-to-apollo-economist/)
[Apollo, The Daily Spark: AI Lowers Wages But Doesn't Cut Jobs (July 30, 2026)](https://www.apollo.com/institutional/insights-news/insights/daily-spark/ai-lowers-wages-but-doesnt-cut-jobs)
[Apollo white paper, Edlich and Slok: The Impact of AI on the U.S. Labor Market (July 2026)](https://www.apollo.com/content/dam/apolloaem/pdf/daily-spark/2026//jul/30/Whitepaper-Impact%20of%20AI%20on%20U.S.%20Labor%20Market-2026-R2%201.pdf)
[NBER Working Paper 34984: Artificial Intelligence, Productivity, and the Workforce: Evidence from Corporate Executives (March 2026)](https://nber.org/papers/w34984)
[Reuters via Investing.com: US growth boosted by AI, but no sign of a turning point higher (Oct. 26, 2026)](https://www.investing.com/news/economy-news/us-growth-boosted-by-ai-but-no-sign-of-a-turning-point-higher-4971872)