Why the S&P 493 Haven't Seen an AI Payoff Yet

I read Torsten Slok’s note this weekend and my first reaction was, yeah, that tracks. But I have a different take and it’s that the problem isn’t what he thinks and it will happen regardless of how long it takes.

Slok is the chief economist at Apollo. His Daily Spark from October 24 says AI “has delivered zero financial payoff to corporate America outside of tech.” He backs it with five charts. Revenue per worker, adjusted for inflation, has climbed for the Magnificent 7 since 2023 and has gone basically nowhere for the other 493 companies in the S&P 500. Profit margins show the same split: the Mag 7 near 30%, the 493 hanging around 10%, where they’ve been for years. He also shows no sign yet of AI lifting productivity growth across the economy.

I run technology and marketing for a manufacturer and spend a lot of my time talking with executives at traditional companies. I think he’s right about the numbers. Where I disagree is on what they mean.

Where I agree with Slok

The 493 aren’t seeing a payoff yet. I don’t think anybody on the inside would argue that.

A couple of things are worth saying about the charts, though. They show the 493 as one group. Nothing in them separates the companies that went hard at AI from the ones that bought a few Copilot licenses and called it a strategy. If a few dozen companies in that group are pulling ahead, an aggregate line won’t show it for a while. And progress compounds in business.

Slok also makes a point I agree with completely. He says AI may be a great calculator, but in business intelligence is mostly judgment, and judgment depends on context. AI doesn’t understand accountability, trust, culture or unwritten rules. That’s right. I’d just add that most of that context is exactly what’s locked up right now, in people’s heads and in systems nobody can easily query. That’s the problem companies are working on, and it’s slow and unsexy work. The opposite of what we hear about AI.

He also says mainstream corporate America “does not need the latest model and is unwilling to pay for it.” I read that differently. Of course they won’t pay for a better model yet. Most of them haven’t built the plumbing that would let them use the one they already have. But there’s a big difference between using frontier models for very specific purposes, like coding agents, and using them for day-to-day deployed work. Any company that’s doing this well right now is using a classifier model like Jev, Open APIs or Decisions API instead of something on the real frontier.

Why big companies are slow

If you’ve never worked inside a big company, it’s hard to describe how long simple things take.

The data is the first wall. It lives in an ERP that’s been customized over 20 years, a CRM nobody fully trusts, a pile of spreadsheets and a few systems only one person knows how to run. Getting an AI agent clean access to any of that means security reviews, vendor contracts, data cleanup and debate over who owns what.

Then there’s the approval chain. Every change goes through reviews: IT, security, legal, compliance, sometimes a committee that meets once a month or even once a quarter. None of that is stupid. Those processes exist because something went wrong once upon a time. But they were built for software you buy once and roll out over a year, not for tools that change every few weeks.

And very little of it moves unless leadership drives it from the top. In the companies I’ve seen make real progress, the CEO or C-Suite is the one pushing it, personally, and the approvals get faster because of it.

There’s one more piece I see up close. Individual people are getting a lot more done with these tools. But that output doesn’t roll up yet. The manager one level up often doesn’t have the context for what got built, what decisions were made along the way, and why they were the right calls. That takes time and Wall Street isn’t the most patient bunch.

What executives tell me

I talk with a lot of executives in traditional industries: finance, agriculture, manufacturing, even a few in space. When AI comes up, the story is almost always the same. They’re struggling (STILL!) to get their engineering teams to work with coding agents at all, much less build AI systems that run inside their own environments. Most leaders literally do not know what is possible and it gets harder to keep up with every day.

That’s what I hear. And it’s consistent enough that I believe it. These companies are paying attention to AI. They grasp that it will drive shareholder value. But they’re stuck between a board asking what the AI plan is and an organization that can’t get a pilot through security and compliance.

2026 looks like 1997 all over again

This feels a lot like the internet around 1997 to me.

In 1997 the technology clearly worked. Amazon was selling books online. But most companies had a brochure website and no idea what to do next. The infrastructure that made e-commerce a real business didn’t exist yet: payment processing you could trust, shipping integrations, ad platforms, analytics, the people who knew how to run all of it. That got built over the next decade, mostly from scratch, mostly by people figuring it out as they went.

Economists have seen this pattern before. Robert Solow wrote in 1987 that “you can see the computer age everywhere but in the productivity statistics.” It took years after that for the gains from computers to show up in the numbers.

That’s where we are with AI inside the enterprise. The models are good. Really good. The infrastructure to use them inside a company with real security, real data and real accountability is being built right now but it’s slow going. And in my view, the part that matters most for the 493 (agents that can actually do work inside a company’s own systems, without someone babysitting every step) has really come together in the last three months and will begin showing in the numbers in 2027. But to this point, that’s not much time for anything to show up in quarterly margins.

Similar to the internet transformation from 1997 – 2003 (and the mobile one in 2007+), we’ll look back on this in a few years and realize it was kind of a mess, and that there were whole new ways of working we didn’t know existed yet.

Slok says the AI story will be decided in the S&P 493. I agree with him. Change in big companies and in society takes time. Data is locked behind security and bureaucracy, and people resist change. That doesn’t mean AI isn’t transformational. It absolutely is, and it’s still coming faster than you think.

The thing I’m not sure the S&P 500 sees coming is the one-person and ten-person companies. They don’t have all the bureaucracy, security, and compliance concerns of big companies, and they don’t have anything to lose. They’re the ones that are going to leverage this technology the most quickly and make a run up into the S&P much more quickly than anyone would have thought.

Published October 26, 2026. Written as the long-form companion to my LinkedIn carousel on Apollo’s October 24 Daily Spark; charts and quotes checked against the note.