What 89% of Small Businesses Using AI Are Getting Wrong

Nearly 89% of small business owners say someone in their company uses AI. That number is real: it comes from a 2025 survey of 3,700 small business owners run by ICIC, a small-business research group, and funded by Intuit. It is also, on its own, a nearly useless signal for whether AI is actually doing anything for the business. If you are asking "everyone says they use AI, why haven't we seen results," the 89% figure is part of the reason you feel stuck: it counts an occasional ChatGPT prompt the same way it counts a fully running automation, and for most companies it is measuring the first thing, not the second.

See how we check whether your data is actually ready for AI before you build anything on top of it.

Why does 89% adoption feel like it hasn't changed anything?

Because for most businesses, it hasn't. In the same ICIC and Intuit survey that produced the 89% number, the leading use cases were data analysis (26.8% of owners), creating marketing materials (25%), and drafting emails or other communication (23.4%). Those are individual tasks a person does faster with help, not workflows AI runs on its own. More than 60% of owners in that survey did report a productivity boost, which is genuine, but a faster email is not the same thing as a business that runs differently because of AI. Adoption measures whether anyone touched an AI tool. It says nothing about whether AI changed how the work actually gets done.

This distinction is not a technicality. It is the entire reason two business owners can both say "we use AI" and mean completely different things: one means a staffer occasionally pastes a paragraph into ChatGPT, the other means leads get qualified, routed, and followed up automatically before a human ever looks at them. Both count in an adoption survey. Only one produces a result you would notice on a P&L.

What does "using AI" actually mean in these surveys?

It usually means "at least one person at the company tried an AI tool at least once." That is a low bar, and it is why adoption numbers across recent surveys cluster so high: Goldman Sachs' 10,000 Small Businesses program found 76% of small businesses currently use AI. Salesforce's SMB Trends Report put "experimenting with or using AI" at 75%. Bluevine's 2026 small business report found 74% using or testing AI tools. All of these numbers, 74% to 89% depending on the survey, are asking some version of "has anyone here used AI," and they all land in roughly the same wide, high range.

The problem is that "has anyone used AI" and "does AI run our business better" are different questions with very different answers, and most coverage of these surveys reports the first number as if it answered the second.

Why does the real number look so different when you ask a stricter question?

Because when a survey asks about depth instead of exposure, the numbers collapse. Three independent sources confirm the same shape:

The starkest evidence comes from the U.S. Census Bureau's Business Trends and Outlook Survey, a government-run, statistically representative survey rather than an opt-in marketing poll. It originally asked whether a business used AI "to produce goods or services," then broadened the question to whether AI is used in any of 15 business functions, from finance to customer service. Even under that wider definition, fewer than 20% of firms with four or fewer employees reported using AI in any business function as of the survey window through May 2026. Firms with 100 to 249 employees came in at 32%, and firms with 250 or more employees at 37%. Adoption is not just lower under a strict definition, it is dramatically lower, and it is lowest exactly where the smallest businesses sit.

Both numbers, the 89% and the sub-20%, are correct. They are measuring different things: whether anyone tried AI, versus whether AI is doing real work inside the business. A buyer trying to make sense of "everyone uses AI" headlines against their own lack of results is running into this exact gap, not a personal failure to keep up.

What's the actual gap between an AI tool and an AI automation?

An AI tool is something a person opens on purpose and prompts by hand, every single time: typing a request into ChatGPT, pasting text into a summarizer, asking a chatbot to draft a reply. It requires a human to start it, drive it, and use the output. An AI automation is a defined workflow: a trigger happens (a lead fills out a form, an invoice lands in an inbox, a support ticket opens), and AI carries the process through to a finished, useful outcome without someone re-prompting it at every step.

The 89% figure is almost entirely made of the first category. That is not a criticism, individual AI tool use is genuinely useful and worth keeping. But it means most of the "89%" have not yet done the harder, more valuable thing: identifying a real workflow, making sure the data behind it is clean enough to act on, and wiring AI into that workflow so it runs without a person carrying it by hand each time. That second step is where the actual business result lives, and it is also where most companies stall.

What are small businesses actually using AI for right now?

Mostly generic, individual tasks rather than embedded workflows. Per the ICIC and Intuit survey, the top three use cases were:

• Data analysis (26.8% of owners): pulling insight from a spreadsheet or report, one time, by hand
• Creating marketing materials (25%): drafting a post, an ad, or a caption on demand
• Drafting emails and communication (23.4%): writing a message a person then reviews and sends

Usage also varies enormously by industry: 63.7% of real estate business owners reported using AI, versus 17.8% in the information industry, showing that "small business AI adoption" is not one uniform behavior but a patchwork that depends heavily on the type of work. None of the top three use cases above describe a workflow that runs itself. They describe a person using a faster tool for a task they were already doing by hand, which is a real productivity gain but not the transformation the 89% headline implies.

Why do the smallest businesses adopt AI slower, even though they need the leverage most?

Time, mostly, followed by not knowing what is actually possible. In the ICIC and Intuit survey, 41.3% of solopreneurs and 38.4% of businesses with one to five employees cited lack of time as a barrier to using AI more, compared to 29.6% at businesses with 21 to 50 employees. That gap narrows steadily as businesses get bigger, which makes sense: a five-person company does not have a person whose job is to figure out AI, while a fifty-person company might.

The next largest barrier was not knowing enough about the tools: 71.9% of respondents across the survey cited insufficient knowledge as an obstacle. And 33.6% said AI tools were "not useful for their business," a genuine perception-of-relevance problem, not just a skills gap. Less than half of business owners across every industry in the survey expressed high confidence in AI. Put together, the smallest businesses face the tightest time constraints, the least internal expertise, and the most doubt about relevance, which is exactly the combination that keeps someone at "I tried ChatGPT once" instead of "we have an automation running."

How do you close the gap: from "we use AI" to "we have an AI automation"?

Four steps, in order, and skipping the first is the single most common mistake.

1. Pick one workflow with a clear trigger and a measurable outcome. Not "use more AI," a specific process: every inbound lead gets qualified and routed, every invoice gets read and matched, every support ticket gets triaged. If you cannot name the trigger and the finished state, you do not have a workflow yet, you have an aspiration.
2. Check whether the data behind it is actually usable. AI acting on messy, incomplete, or scattered data produces messy, incomplete, or wrong outcomes faster than a person would. This is the step almost everyone skips, and it is why so many pilots quietly die: the workflow was fine, the data underneath it was not ready.
3. Build the automation to run without a person re-prompting it. The output of step one is a live process, not a saved prompt. If a human still has to open a tool and manually trigger every run, you have made tool use faster, not built an automation.
4. Measure the one outcome you defined, then add the next workflow. Resist the urge to roll out AI everywhere at once. One working automation, with a number attached to it, earns the trust and the budget to build the second one.

This is deliberately narrow. The businesses stuck at 89%-style tool use usually got there by trying to "adopt AI" broadly instead of automating one real thing completely.

What mistakes keep businesses stuck at the tool-use stage?

Three show up constantly, and each maps to a step above.

• Treating "someone tried it" as done. A staffer using ChatGPT for emails is a fine start, but it is not a strategy, and it will not show up as a business result no matter how often it happens.
• Skipping the data check. Businesses that jump straight to building an automation on data that was never cleaned or connected end up with an unreliable process that gets abandoned within a few months, then get counted as "we tried AI and it didn't work."
• Spreading effort across everything at once. Trying to bolt AI onto five processes simultaneously, instead of finishing one, is why so many companies report using AI while also reporting no measurable change: nothing gets far enough to actually finish.

None of these are AI failures. They are sequencing mistakes, and every one of them is fixable before you write a line of anything.

Two businesses, both counted in the 89%: what actually separates them?

Picture two ten-person companies, both of which would answer "yes, we use AI" on a survey like ICIC's.

The first has a marketing coordinator who pastes product notes into ChatGPT to draft social captions, and a bookkeeper who occasionally asks an AI tool to summarize a report. Both are genuinely useful habits. Neither changes what happens if that person is out sick, neither shows up in a revenue or cost number anywhere, and neither would survive being described to an investor as "our AI strategy." This company is real, common, and squarely inside the 89%.

The second company has one workflow rebuilt end to end: every inbound lead is enriched, scored against a defined ideal customer profile, and routed to the right rep automatically, with a human only stepping in for the deals that clear a value threshold. Nobody has to remember to run it. It is also inside the 89%, technically, but it is a fundamentally different business from the first one. The difference between them was never willingness to try AI. Both tried it. The difference was whether one specific, real workflow got finished, measured, and left running.

That is the self-check worth applying to your own business before spending another dollar on "more AI." Not "do we use AI," which nearly everyone will answer yes to, but "name one workflow, right now, that finishes without a person driving every step." If you cannot name one, you are the first company, not the second, and the fix is not more tools. It is finishing one real automation.

If you want a second set of eyes on which workflow to pick first and whether your data can actually support it, that is exactly what our AI feasibility and data readiness work does: an honest check before you build, so the first automation you ship is one that sticks instead of one more tool nobody opens twice. From there we can plan, build, and run the automation with you. Book a free consultation below and we will walk through your first candidate workflow together.