Apollo Global Management's chief economist, Torsten Slok, said in an October 24 note that AI has delivered "zero financial payoff to corporate America outside of tech." He split the S&P 500 into two groups: the seven biggest tech names, and everyone else. Only the first group's profit margins are actually moving. If you've spent money on AI this year and can't point to the AI ROI it produced, that puts you in the majority, not the exception. The real question is whether that's an AI problem or a measurement problem.
What Did Apollo's Economist Say About AI ROI?#
Slok calls the second group the "S&P 493," the S&P 500 minus Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla. Per TechCrunch's October 25 report on the note, that group's margins and revenue per worker are flat, while the Mag 7's margins have climbed from roughly 11% in 2015 to close to 30% now. His line: "the AI story will be decided in the S&P 493, not the Mag 7." Alphabet is one of the names funding that climb, and its latest model run is a story of its own: Gemini 4 Argon reportedly processed almost 25 trillion tokens in 72 hours this October.
At WebEpex we've watched this exact split play out in miniature with GCC and European automation clients this year. The ones who treat AI spend as a line item with a number attached move something. The ones who treat it as a mood board don't.
Does This Apply to a Small Business, or Just the S&P 500?#
If you've added one AI tool to one workflow and you're genuinely tracking what changed, this piece isn't about you. You're already doing the thing Slok says most companies skip. If you've added three or four AI subscriptions this year because they were cheap and everyone else was doing it, and couldn't say in a sentence what each one moved, that's different, and it's the same pattern we flagged when SaaS point solutions started getting cut four times faster for the same underlying reason: nobody could prove the tool was earning its seat.
A Basis survey cited alongside Slok's note found that 88.6% of 184 organizations pay for premium AI tools, and only 6.5% measure the return against a defined benchmark. Scale doesn't change that ratio much. It's the same habit at $200 a month or $2 million.
A few more numbers worth sitting with, all cited in that same report:
- KPMG surveyed 2,145 senior leaders: 49% scaled back, narrowed, or paused AI agent rollouts; only 7% reported established ROI.
- TransUnion found 53% of enterprise marketers reported "meaningful AI ROI" as of August, mostly from time saved, not revenue moved.
- Meta and YouGov surveyed 3,983 small business owners across Asia-Pacific: 76% reported a productivity gain, self-assessed, with no benchmark attached.
Everyone's measuring something different here. Almost nobody's measuring the same thing twice, which is exactly why the averages come out flat.
The Honest Trade-Offs in Slok's Argument#
I think he's right about the measurement gap and overreaching a little on the cause. His chart shows timing, not causation. Flat S&P 493 margins could reflect pricing power, cost discipline, or simple cyclical recovery rather than AI specifically, and a company-wide average says nothing about the individual business down the hall that actually rewired one workflow, tracked the number, and watched it move. Slok's own line is the tell: "AI may be a great calculator, but in business, intelligence is mostly judgment." That's a real argument against buying tools and hoping. It's not an argument against AI.
How We're Already Building Around This#
We used to scope automation work around the tool first and the metric second. Wrong order, and it took two stalled client builds to see why, because nobody could tell afterward whether the thing had actually worked. At WebEpex every automation build now starts with naming the one number it has to move before a line of n8n gets wired up: a WhatsApp lead-qualification flow gets judged on response time and qualified-lead rate, not on whether the bot sounds clever.
We reviewed Slok's note the day it ran, and it matched something already priced into how we scope work. GCC, European, US, Canadian, and Indian SMB clients who ask for "AI" without a metric attached rarely come back for a second project, because nobody can agree afterward whether the first one worked, the same layoffs-to-AI-pivot logic HubSpot's own restructuring ran into when the bet wasn't tied to a tracked number either. That's the tell.
What I'd Tell a Client Asking About This#
Before you spend on anything AI-labeled this quarter, write down the one number it's supposed to move, check it again in 30 days, and be honest if it didn't budge. If you can't name that number right now, don't spend yet. That's not caution for its own sake: it's the exact gap Slok's data is describing, just applied to your business instead of the S&P 500. And if the honest answer is "I don't know what to track," that's a smaller, cheaper problem to fix first.
If you want a second opinion on whether something you're running, or about to buy, actually has a benchmark worth tracking, send me what it is and I'll tell you straight, no pitch attached. cal.com/webepex/growth-review
