Why AI Is Paying Off for Tech but Not Everyone Else, and How to Change That

AI is paying off for the tech sector. For almost everyone else, it hasn't shown up in the numbers yet. That's not a sign AI doesn't work outside tech. It's a sign that most businesses are using it the wrong way.

What the data says

On October 24, 2026, Torsten Slok, Chief Economist at Apollo, published a short note titled No Signs of AI Payoff Outside the Tech Sector. He compared two groups of US companies:

• The Magnificent 7: the seven largest tech companies, such as Microsoft, Alphabet and Nvidia.
• The S&P 493: the other 493 companies in the S&P 500. Banks, manufacturers, retailers, insurers, logistics firms. In other words, most of the real economy.

His findings:

• Revenue per worker is rising for the big tech companies. For the other 493, it is flat (adjusted for inflation since late 2019).
• Profit margins are rising for big tech and flat for everyone else.
• Productivity growth across the economy shows no AI boost yet.
• AI spending is concentrated in a small number of firms.
• Mainstream companies are pulling back. With nothing reaching the bottom line, many now see their AI usage bill as a cost that eats into margins, not a source of growth.

His conclusion: the AI story will be decided by the 493, not the Magnificent 7.

We agree with every word. We see the same pattern in the businesses we work with. The question is not whether AI can pay off outside tech. It's how.

This article explains why the gap exists, and the step-by-step process that closes it.

Why tech gets the payoff first

Tech companies aren't smarter about AI. Their work is simply easier for AI to do. Two things make the difference.

1. Their work comes with its context already written down

Think about what an AI coding agent has access to:

• The full history of every change, in version control
• Requirements and tasks in a ticketing system like Jira
• Design and architecture documents
• Documentation for every tool, one search away

Everything the agent needs to do the job is in one place, in a form it can read.

2. Their work is easy to check

Code either passes its tests or it doesn't. A software team gets a free, instant way to check whether the AI did the job right. When checking is cheap, you can let AI do more on its own, and mistakes get caught before they cost anything.

Software gets a built-in way to check AI's work for free. Almost every other business has to build one.

That's why the gains show up first in tech. It's not the industry label. It's the shape of the work.

Why everyone else can't simply copy it

Now look at how work runs in an insurer, a logistics company or a manufacturer. Take an insurance claim, a supplier invoice or a customer complaint.

• Customer details sit in the CRM. Transactions sit in the ERP.
• Conversations happen over email, phone calls and WhatsApp.
• Policies and SOPs live in scattered documents, often out of date.
• When to escalate, and to whom, lives in someone's head.

The AI isn't missing intelligence. It's missing context.

Slok makes a related point. He notes that in business, intelligence is mostly judgment, and judgment depends on context. He lists the things AI doesn't understand on its own: accountability, trust, culture, unwritten rules, ethical trade-offs and fairness.

We agree, and this is exactly where the work is. Those unwritten rules are what separate your best employee from a new hire. Until they are captured and given to the AI, it will behave like a new hire on day one. Fast, but not trusted.

Ask yourself: if a new employee joined tomorrow, how much of what they need to know is actually written down?

There is also no free way to check the work. A claim decision doesn't come with a test suite. A business has to design its own checks: sample reviews, approval steps, clear rules for when a person must sign off. That takes deliberate effort, and it's one reason progress is slower outside tech.

Why buying AI tools hasn't moved the numbers

Most businesses started their AI journey the same way. They bought licences for ChatGPT, Claude or Copilot and handed them to staff. That was a good first step. It just isn't the last one.

Here's why it doesn't show up in revenue per worker or margins:

• The gains stay personal. One person writes an email three times faster. The business process around that email doesn't change.
• Results depend on who is using the tool. Curious, skilled people get a lot out of it. Others get little. You can't standardize what lives in individual chat windows.
• Speeding up one step doesn't speed up the whole process. If a quote takes five days to go out, and AI saves an hour on one step, the customer still waits five days. The delays sit in the handoffs, approvals and waiting between steps.

Adoption is people using AI. Transformation is the business running differently because of it. Only the second one shows up on the balance sheet.

We've seen this before. In the 1990s, Microsoft Office made every individual faster. But the big business gains came later, from ERP and CRM systems, where companies redesigned and standardized their processes. AI is in its Microsoft Office moment right now. The ERP moment is where the ROI is.

Why AI spending feels like a cost, not an investment

Traditional software is mostly a fixed cost. You pay for it, and running it costs little more. AI is different. Every task an AI model performs uses "tokens", and every token is billed. Think of it as an electricity bill for intelligence. It never stops running.

Slok observes that mainstream companies don't need the newest AI model and aren't willing to pay for it. We'd go further: most business tasks shouldn't use the newest model at all.

The common mistakes:

• Using the most powerful model for every task. Reading an invoice doesn't need the same model as drafting a legal argument.
• Measuring usage instead of outcomes. A rising AI bill tells you people are busy, not that anything improved.
• Finding the cost after deployment. The bill arrives before anyone has checked whether the use case was worth it.

The bill only feels like a cost when nothing on the other side of it changes. Tie every AI dollar to a process you can measure.

The fix is simple to state: match the model to the task, design the work so the AI only sees what it needs, and check the business case before going live, not after the first invoice.

How to change that: a process, not a switch

There is no plan you can switch on and no tool you can buy that does this for you. There is a process. Businesses that follow it get results. Businesses that skip steps get pilots that never scale.

1. Start with a process, not a tool. Pick one workflow with clear volume, cost or delay. Invoice processing, quote preparation, claims review, order entry. Measure how it runs today: time, cost, error rate.
2. Capture what your best people know. Write down the rules, exceptions and judgment calls your experienced staff carry in their heads. When to escalate. Who to ask. What "good" looks like. This is the context the AI is missing.
3. Start where the work is easy to check. Let AI fully handle the steps you can check automatically, like matching an invoice to a purchase order. Let it draft, and a person approve, where expert judgment is needed. Keep people in charge where there is no clear right answer.
4. Redesign the work for people and AI together. Decide which steps AI agents handle, which steps people handle, and where a person must approve before anything moves. Bring the scattered context into one place the agents can use.
5. Prove it, then scale it. Compare the new process against the old numbers. Move to the next process only once the first one delivers.

Every step here is about your business, not about AI. That's the point. The technology is ready. The business usually isn't, yet.

This is the work Slok's numbers are waiting for. It's also why the payoff takes longer outside tech: steps 2 and 3 have to be built, where software companies get them for free.

What realistic ROI looks like outside tech

Don't expect what the Magnificent 7 are seeing, and don't expect it in a quarter. Outside tech, AI ROI looks different:

That isn't a weaker outcome. It's a more durable one. Each process you redesign keeps paying back. The know-how you capture stays with the business when people leave. And each new process is faster to tackle than the last, because the groundwork is already there.

The goal isn't overnight transformation. It's one process that measurably runs better, then the next one.

Here is what that looks like in practice. Life Insurance Direct has compared life cover for Australians since 2006, and almost all of it happens on the phone. Their QA team used to check 2 to 5% of calls by hand, and feedback reached advisers 3 to 4 weeks late. We built Metricsense to read and score every call, then rank them by risk. Now every call is checked, the team reviews in risk order, and QA has scaled with no added headcount. A person still confirms every flag.

What we've been saying from the field

Slok's data is new. The pattern isn't. Over the past few months, I've written about each piece of this gap, based on what we see working with businesses on AI.

How Avesta Labs helps

We work with businesses outside the tech sector to do exactly this: one process at a time, with people in control.

• AI Kickoff: we map your processes, find the ones most ready for AI, and capture the know-how your agents will need. You leave with a clear, measured starting point.
• Avesta AgentOS: the platform where your data, knowledge, policies, people and AI agents work together on real business processes. Every agent action is logged, and people approve what matters.
• Pathway Foundations: if your teams are still at the personal AI stage, we help them use those tools well and consistently first.

If you're one of the 493, and your AI spend hasn't shown up in your results yet, talk to us. We'll start with one process.

Sources

• Torsten Slok, No Signs of AI Payoff Outside the Tech Sector, Apollo Daily Spark, October 24, 2026