The bubble may pop. The tools are staying.

Let’s start by agreeing on something, because it will save us both time: a lot of what you’ve been told about AI is nonsense.

Not “slightly exaggerated.” Nonsense. The demo that turns out to be edited. The startup with no customers and a $4 billion valuation. The LinkedIn post claiming a chatbot replaced an entire department. The consultant who wants $40,000 to install a tool you could have set up yourself in an afternoon — if anyone had bothered to tell you which afternoon.

If your reaction to all of that has been a quiet “sure it did,” you have good instincts. Keep them. You’re going to need them, because the useful part of this technology is buried under an enormous amount of noise, and separating the two is a skill. It’s the skill, actually. And the people who have it are almost never the ones shouting about AI on the internet.

So here’s the argument, up front: the AI market is almost certainly overheated, the correction is probably coming, and none of that means AI is going away. Those two things have coexisted before. They’re about to coexist again. And the businesses that come out the other side ahead will be the ones that spent the noisy years quietly figuring out which two or three things actually worked for them.

Part One

Yes, it’s a bubble. The numbers aren’t subtle.

We’re not going to talk you out of the bubble thesis, because we mostly agree with it. The spending and the returns have come unglued from each other, and you don’t need a finance background to see it.

That last number is the one worth sitting with. Depending on which survey you read, small-business AI adoption is either near-universal or barely underway — and both are honest, because they’re measuring completely different things. Ask “have you ever used ChatGPT to write an email” and you’ll get a number like 58% from the U.S. Chamber of Commerce. Ask “does AI touch how you actually produce your product” and it drops to under nine percent.

The gap between those two numbers is where all the hype lives. It’s also where all the opportunity lives, which is an annoying thing about hype — it tends to cluster around things that are genuinely valuable, which is why it’s so hard to dismiss cleanly.

Part Two

We have done this before, and it went fine

In March 2000, the dot-com bubble peaked. Over the next two and a half years the NASDAQ fell more than 75%, and something on the order of $5 trillion in market value evaporated. Pets.com became a punchline. Webvan burned through more than a billion dollars. Serious people wrote serious columns about how the internet had been oversold.

And then everyone kept using the internet.

Because here’s what the crash actually killed: business models that never made sense. It did not kill e-commerce, or search, or email, or the idea that a small business should probably have a website. Amazon’s stock lost roughly 90% of its value — and Amazon used the bust to build the infrastructure it still runs on. The technology survived. The nonsense didn’t.

The relevant question for your business was never “is the stock market overvaluing this.” It was “does this do something useful for me on a Tuesday.” In 2001, the answer for a plumbing contractor in Kalamazoo was: yes, eventually, you’re going to want a way for customers to find you online. That was true whether the NASDAQ was up or down, and the people who figured it out in the quiet years after the crash did better than the ones who waited for permission.

Same thing now. Whatever happens to AI valuations in 2027, the software that drafts your invoice reminders will still draft your invoice reminders. It costs about what a phone line costs. Its usefulness has nothing to do with anyone’s stock price.

Part Three

Why the skeptic is the best person in the room

Go back to that MIT figure — 95% of pilots delivering nothing. That’s a genuinely damning number, and the AI boosters have never had a good answer for it. But read past the headline and the finding gets more interesting, because MIT’s researchers were clear that the failures weren’t technical. The models worked. What failed was the organizational part: companies bought tools and never changed a single workflow around them.

Which is to say: the 95% failed for exactly the reasons a skeptic would have predicted.

They bought first and asked questions later. They started from “we need an AI strategy” instead of “invoicing takes eleven hours a week and I hate it.” They let a vendor define success. They ran a flashy pilot in a department that wasn’t struggling, declared victory, and quietly stopped using it four months later.

The person who says “prove it” before spending money is not the obstacle in this story. They’re the control group. And there’s evidence that this disposition pays: firms that actually train their people on the tools — even just a handful of hours each — report substantially better results than firms that buy licenses and hope, with some analyses putting the gap at more than double. Skepticism is what forces that training conversation to happen.

Part Four

What “a real tool for a real problem” actually looks like

The reason AI feels like vapor to a lot of business owners is that the examples are always vapor. “Transform your customer journey.” “Unlock insights.” Nobody has ever had a Tuesday problem called “my insights are locked.”

Here’s the kind of thing we mean instead. None of it is impressive. All of it is hours.

Notice what’s missing from all three: nobody got replaced, nothing got “transformed,” and no one had to believe anything. The work that disappeared is work no human ever wanted — the re-typing, the chasing, the copying between two systems that should have been talking all along.

That’s the whole pitch. It’s deliberately unglamorous. Unglamorous is the point — it’s the part that survives the correction.

Part Five

The cost of waiting isn’t dramatic. That’s why it’s dangerous.

Nothing bad happens to you next quarter if you sit this out. That’s the trap. There’s no cliff, no moment where a competitor obviously eats your lunch. There’s just a slow divergence: the shop across town starts answering inquiries in twenty minutes instead of two days, and their books close on the third of the month, and a year later they’re bidding on work you don’t have the capacity to touch.

The other cost is fluency. Technology landscapes don’t send a notification when they shift — and the businesses that fell behind on the internet, on mobile, on cloud accounting didn’t decide to fall behind. They just kept meaning to look into it. Two years of not-looking is how a manageable learning curve turns into an intimidating one, and intimidation is what makes people overpay later, in a panic, for something they don’t understand.

You don’t have to move fast. You do have to stay in the conversation.

Staying in the conversation is cheap. It costs a few hours a quarter and a willingness to test one small thing at a time. Panicking in 2028 because you’re four years behind is expensive.

Part Six

Where KalFlow fits

We’re two operators from Southwest Michigan with more than 30 years of combined experience inside growing businesses — running operations, closing books, and answering the customer. We’ve been the person the broken process lands on, which is why we’re allergic to the version of this conversation that starts with software.

We don’t sell you a platform and disappear. We map how your work actually moves, find the two or three places where automation pays for itself quickly, build those, and document everything so your team can run it without us. Our fractional services cover the four functions where the busywork concentrates:

• Operations — workflow mapping, SOPs, and putting the repeatable parts on rails.
• Accounting & finance — AR/AP automation, a month-end close that doesn’t eat a week, cash-flow reporting you can read on a phone.
• Marketing — content pipelines that keep running when the week gets busy, and lead routing that reaches a human before the lead goes cold.
• Customer service — intake and triage, AI-drafted replies your team reviews and sends, still in your voice.

If what you’d rather have is your own people getting confident with these tools, that’s a different engagement and an equally good one. Our customized company workshops are on-site, hands-on, and built around your actual processes — not a generic slide deck about prompt engineering. We also run open community sessions through KalFlow Events, where the price of admission is showing up, and skeptical questions are the ones we like best. Between those, Keeping Up with KalFlow is where we publish what’s actually changed and what it means for a business your size.

In closing

Keep the skepticism. Lose the distance.

The correction, if and when it comes, is going to be loud. Valuations will fall, some well-known companies will fold, and there will be a stretch of months where saying “I told you so” feels great.

Then Monday will arrive, and the invoices will still need chasing.

The tools that solve small, boring, repeated problems don’t depend on anyone’s stock price. They’re already cheap, they’re already here, and they reward the person who tests carefully and refuses to be impressed — which, if you’ve read this far, is probably you.

Bring the doubt. We’ll bring the process map.