Article · AI, and what it’s actually doing
AI pays back in speed before it pays in profit
Apollo's chief economist finds AI has lifted profit only for the 7 tech giants. In agencies the payback is real, but it shows up as speed, not margin.
Torsten Slok, the chief economist at Apollo, published a note on 24 October with a blunt title: no signs of AI payoff outside the tech sector. He split the S&P 500 in two. The seven giants, Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla, earn about $780,000 of revenue per worker at margins close to 30%. The other 493 companies sit at about $525,000 per worker and 10% margins, and neither figure has moved since 2019 once you adjust for inflation. Cryptopolitan has the figures.
His line is that AI “has delivered zero financial payoff to corporate America outside of tech.” Spending data from Ramp, the corporate card company, shows how lopsided it is. The top 10% of firms account for 99.5% of the money spent running AI models. Most ordinary companies, Slok says, now treat the AI bill as a cost that eats into margin, and they are pulling back.
What doesn’t the chart see?
I don’t doubt the numbers. I think they measure the wrong thing, or at least measure it too early.
Where AI is paying back for the businesses I see is speed. At Embryo we turn proposals and pitches round much faster than we ever have, and some of the work we now deliver in days would have taken weeks to put together not long ago. That doesn’t show up as revenue per worker in the quarter it happens. It shows up as more pitches entered, more of them answered properly, and a new business pipeline that moves faster than it used to.
It’s the same with hiring. When a team can take on more work without adding a role, the saving is a recruitment cost that never happened. Nobody books an invoice that wasn’t sent, so it’s real money that never appears on a chart of margins.
Why does Slok still have a point?
Efficiency isn’t profit until somebody decides what to do with the time. If a proposal takes days instead of weeks and the team fills the gap with more meetings, the business is no better off. That’s the trap in the AI pilot that never lands: the tool works, the old job carries on, and the saving evaporates.
It’s also why finance teams have started capping AI spending, as Accenture did with Claude. If the gain is speed and nobody can point to where the speed went, a monthly bill for AI looks like pure cost.
Slok’s own conclusion is the useful one: “The AI story will be decided in the S&P 493, not the Mag 7.” Most businesses, and nearly every agency, are in the 493.
How should you measure it?
Measure the thing AI is actually changing. Time from brief to proposal. Pitches answered each month. Roles you didn’t need to hire. Then decide in advance what the freed-up time is for, whether that’s more new business, better work for the clients you have, or fewer late nights.
If you can’t say where the speed went, you’ll end up agreeing with Slok’s chart, and the budget will go to whoever can.