One quarter. That's how long it took for the "we're interested in AI but haven't started yet" crowd to basically disappear.
Pax8's Q2 2026 SMB AI Pulse Report came out July 13. Between Q1 and Q2, the share of small businesses saying they're interested but haven't started fell from 9% to 1.5%. That's a survey of 402 U.S. small business leaders, companies between 5 and 499 employees.
For years, every AI article aimed at small businesses had the same message: adopt or get left behind. That argument is dead. At 1.5%, everyone has adopted something. The race to start is over, and nobody won anything by starting.
Because here's the number that actually matters: 28.9% of those same businesses are stuck in experimentation. They bought the tools. They ran the trials. Nothing made it into the way the business actually runs. Pax8's own VP of AI Adoption put it plainly: almost one in three small businesses is stuck in experimentation right now.
We'd call that pilot purgatory. You're not behind, but you're not moving either. You're paying for AI, you'd tell a survey you "use AI," and your Tuesday looks exactly like it did last year.
The gap between "using AI" and AI doing work
The self-reported numbers have always been generous. Surveys say somewhere between 58% and 82% of businesses have "adopted AI." The U.S. Census Bureau, measuring actual production use as of May 2026, puts it around 17 to 20%.
That spread isn't a rounding error. It's millions of businesses where AI exists as a browser tab, a subscription, and a vague sense of guilt.
And it's not for lack of buying things. SBE Council's 2026 survey found the typical AI-using small business pays for a median of five AI tools. Five. Meanwhile production use sits under 20%. A lot of businesses are funding a stack nobody operates.
So if the stuck third isn't under-tooled, what separates them from the businesses that actually deployed?
Pax8's data points at two things, and neither one is software. Only 11% of experimenters have a documented AI policy. And 68% of experimenters report their leadership is aligned on what AI is for, versus 91% of businesses in active deployment. Pax8 is careful to call that correlation, not proven causation, and that honesty is worth keeping. But the pattern matches everything we've seen: the businesses that deploy decided, in writing, what AI is allowed to do and who answers for it. The businesses that stall are still having that conversation.
Nobody's job was to make it stick
"We tried AI and it didn't stick" is a sentence we hear a lot. Almost every time, the real story underneath is that nobody's job was to make it stick.
A pilot is easy. Someone signs up, plays with it for two weeks, gets a decent result, shows the team. Then the pilot ends and the tool is just sitting there, because turning a cool demo into a workflow that runs every single day is actual work. Someone has to connect it to your real data. Someone has to rewrite the process around it. Someone has to notice when it drifts and fix it. Someone has to train the people who touch it.
That last one matters more than most owners think. Research from LSE and Protiviti found 93% of employees who got AI training use the tools regularly. Without training, it's 57%. The tool didn't change. The ownership did.
MIT's NANDA group studied why generative AI pilots fail, and one finding is worth stealing: pilots built with an external partner reached production about 67% of the time. Internal do-it-yourself builds made it about 33% of the time. The difference isn't smarts. It's that a partner's entire job is getting you to deployed, while your internal champion has a day job that keeps interrupting.
That's what a tech team is for. Not a vendor who sells you seat licenses and disappears. A team accountable for the boring middle part: the integration, the process redesign, the "it broke at 6am and someone noticed" part.
Why this is worth fixing now
The upside isn't theoretical. In that same Pax8 report, 71% of active AI users said AI lets them compete with much larger companies.
We can vouch for that one personally. Kief Studio is two people. We automated our entire daily content pipeline: topic selection, research, writing, quality scoring, video generation, publishing, distribution to six platforms. It runs on timers with no human in the loop. That's our LTFI system doing the work of roles we never hired for, and it's why two people can cover what would normally take a 10 to 14 person team.
None of that came from buying a tool and hoping. Every piece of it exists because making automation stick is literally our job, so we did for ourselves what we do for clients: picked the workflow, built AI into it, and made someone (us) accountable for it running every day.
And the window here is real. The experimenting share grew last quarter, from 25% to 28.9%. The hesitant businesses didn't leap to deployment; they moved into testing and joined the stuck third. The next wave, AI agents, is already repeating the pattern one layer up. The businesses that learn to cross from pilot to production now will do it again with agents while everyone else runs their second round of experiments.
So if you're in the stuck third, the fix isn't tool number six. It's an owner. Someone whose job is to take the one experiment that showed promise and turn it into a workflow your business runs on, with a written policy, trained people, and a name attached to the outcome.
If you don't have that person, that's the actual gap. Not an agency. Not a consultant. Your tech team. That's what we do, and the first conversation is free -- no commitment, no pitch deck, just a straight answer about whether the thing you piloted is worth deploying.
Members at kief.studio also get the companion resource for this post: a short worksheet for figuring out which of your stalled experiments is worth pushing to production first. Signup's free.