This week a chief economist at Apollo, a big investment firm, told the world that AI has delivered "zero financial payoff" to corporate America outside the tech industry. Cue a lot of headlines. And cue a quiet question from every owner who has paid for a chatbot subscription since spring: is it paying off for me?
Good question. Let me tell you what that headline does and doesn't say, and then give you a way to answer it for your own business in about a month.
What the economist actually said
Torsten Slok looked at the S&P 493, which is the S&P 500 minus seven giant tech companies. According to TechCrunch's report, profit margins and revenue per worker in those companies "appear to be flat", and he sees no evidence of a productivity boost. In a June note on Apollo's own site, he wrote that "so far there are no signs of profit margins rising outside the tech sector", and he expects regulated industries like banking, health care and energy to be slow.
His line I liked best: "AI may be a great calculator, but in business, intelligence is mostly judgment."
What it doesn't tell you
Here is the honest part. This is about big listed companies, measured through profit margins across thousands of employees. It says nothing about a bakery, a clinic or a one-person consultancy. A margin number that is flat can hide a team that saved two hours a week on one task and spent them on something else. It can also hide a team that bought five tools nobody opens.
So the headline is a reason to be careful, not a verdict. If big firms can spend heavily and not see it on the bottom line, you can too. The difference is that you can actually find out, because your business is small enough to see.
Why "it feels faster" is not a result
Most of us judge AI by feeling. The draft appeared in ten seconds, so it must be saving time. But a draft you then rewrite for twenty minutes may cost more than writing it yourself. And a tool that answers customer emails quickly but wrongly once a week can cost you one angry customer, which erases a lot of seconds.
The trap is that the quick part is visible and the slow part (checking, fixing, explaining to a customer) is scattered across your day. Nobody adds it up. That is exactly why big companies struggle to see the payoff, and it is the same reason you might.
A 30-day pay-off ledger
You do not need software. A page in a notebook is fine. Pick one task you gave to AI, or want to, and write down three things.
- Before you start (one week of normal life):
- How many minutes the task takes you, per time.
- How often it comes up.
- How often it goes wrong (a late invoice, an error in a quote, a customer asking again).
- During the month, every time you use the tool:
- Minutes the tool needed, including your checking and fixing.
- Anything it got wrong, and what that cost you.
At the end: multiply. Minutes saved per time, times how many times, minus what you spent on the subscription and on fixing mistakes. If you can't point to a clear positive number, the tool is not paying off yet, whatever it feels like.
One tip: count the money too, not just minutes. An hour saved that you spend scrolling your phone is not a payoff. An invoice chased a week sooner that arrives a week sooner is. I wrote about that in why your first AI agent should chase invoices, because a task with a visible price is the easiest one to measure.
Checking the work is part of the cost
Slok's point about judgment lands here. The tool does not know your regular customer will want the discount, or that a certain supplier always sends the wrong file. You do. That review is the human part, and it takes time. If you skip it to make the numbers look better, the numbers are lying to you. I went through what real checking looks like in how to check an AI agent really did the job.
When the answer is "not yet"
Sometimes the ledger says no, and that is a useful result.
- If the task happens a handful of times a month, the setup and checking will rarely pay back. Keep doing it by hand.
- If a mistake is expensive (a wrong price, a wrong tax figure), the review eats the saving. Keep a person on that step.
- If the month is too chaotic to compare, you need a calmer month before you can measure anything.
- One month is a short test. Some gains take longer to show, and I can't tell you from the outside which side of that line your task is on.
I should also say the evidence on all of this is young. One economist's reading of large-company data is a signal, not a law, and I'd treat the optimistic claims from tool vendors with the same caution.
One thing to do this week
Choose one task, today. Write its three "before" numbers on a page and put the page where you will see it. In 30 days, do the sum. If it is positive, keep it and consider the next task. If it isn't, switch the tool off and say "not yet" with a clear conscience.