Apollo chief economist Torsten Slok’s October 24, 2026 commentary argues that AI’s financial benefits remain concentrated in major technology companies. Its comparisons raise a worthwhile business question: how can spending on AI translate into better results? They do not, by themselves, answer whether a particular company’s AI workflow earns a return. [1]
Start with what Apollo actually compares
Slok contrasts rising revenue per worker and profit margins for the Magnificent Seven with flatter measures for the remaining S&P 500 companies, labelled the S&P 493. He also argues that AI has not produced a visible productivity acceleration. These are his interpretations of aggregate evidence, rather than a controlled evaluation of every business using AI. [1]
The comparison groups deserve attention. The S&P 493 is the index excluding seven companies; it is not a clean classification containing every nontechnology business and no technology businesses. An index comparison can highlight an uneven pattern without representing the experience of every small firm, private company or industry.
Revenue per employee and output per hour answer different questions
BLS defines labor productivity as real output divided by hours worked. Its September 3 revised release reported nonfarm business productivity growth of 1.4% at a seasonally adjusted annualized rate in the second quarter of 2026, and 2.2% from a year earlier. Those are different comparisons of the same broad measure. Neither figure isolates an AI contribution. [2]
Revenue per worker relates sales to a headcount. Apollo labels its revenue-per-worker chart as inflation-adjusted to fourth-quarter 2019, so it should not be read as a simple nominal-sales comparison. Product mix, staffing and hours can still affect interpretation. Two firms could have equal revenue per employee while their employees work different hours; a sales-per-person comparison cannot simply substitute for an output-per-hour measure. [1]
Apollo’s productivity-chart note also identifies utilization-adjusted total factor productivity from the San Francisco Fed. Treat that as another measure, rather than silently equating it with revenue per worker or the BLS headline labor-productivity rate. [1]
AI adoption is a separate measurement
The Census Bureau’s May 26 analysis examined BTOS observations from December 14, 2025 through May 3, 2026. It reported AI usage around 17%–20% during that period. These are historical observations, not an October adoption estimate. [3]
Census also explains a question change: in November 2025, wording shifted from AI used in producing goods or services to AI used in any business function. That broader wording can include tasks such as drafting emails. A usage response therefore establishes neither a financial return nor a comparable production transformation. Check the question and observation period before placing adoption percentages beside earnings or productivity charts. [3]
Use a workflow ledger to make a company decision
For an internal decision, start with one repeatable activity and its current performance. Record volume, accepted output, human time and errors before introducing a tool. Then measure the same items after the change, keeping the acceptance rule consistent. Include review time and failed attempts rather than counting only the fastest successful draft.
Consider an illustrative team that saves 20 hours in a month and values the released capacity at $30 an hour. That gives $600 of potential capacity value. If subscriptions and additional review together cost $400, the difference is $200 before other implementation costs. It is not automatically $200 of cash profit: salaried time saved may be redeployed without reducing payroll, and unused capacity may produce no added revenue.
Record what happened to the released hours. Did the team complete more accepted work, reduce overtime, shorten a backlog or simply finish earlier? Keep capacity value, cash savings and incremental margin in separate fields. This makes a later decision to expand, revise or stop the workflow easier to assess.
What would strengthen the evidence?
A stronger company-level assessment would connect actual tool use with comparable outcomes over time, account for implementation costs and explain other changes in the business. Aggregate charts can motivate that investigation; a single successful pilot cannot establish an economy-wide productivity acceleration either. Read Apollo’s warning as a reason to measure results carefully, while keeping the level of each claim matched to its evidence.
AI payoff evidence check
Sources
- [1] Apollo, Torsten Slok, October 24, 2026: https://www.apollo.com/wealth/insights-news/insights/daily-spark/no-signs-of-ai-payoff-outside-the-tech-sector
- [2] BLS, Productivity and Costs, Second Quarter 2026, Revised, September 3, 2026: https://www.bls.gov/news.release/prod2.nr0.htm
- [3] U.S. Census Bureau, May 26, 2026, historical BTOS AI-use analysis and question wording: https://www.census.gov/library/stories/2026/05/ai-use-businesses.html