The 2026 Michigan Entrepreneurship Score Card landed in April with a sentence that should stop any Southwest Michigan owner mid-scroll: the state’s small business economy is “stable but constrained.”

Not collapsing. Not booming. Stuck — with business openings and closings running nearly equal for two straight years, and rising labor, energy, and health care costs quietly deciding which hires don’t happen and which expansions get pushed to next year.

If you run a business here — a storefront a few minutes from the lakeshore, a supplier off the I-94 corridor, a family operation out in orchard country — none of that is news. You’ve felt it in the quote you didn’t chase because nobody had time, in the invoice that went out eleven days late, in the good candidate you couldn’t justify at the number they needed.

Michigan, by the numbers

The AI gap isn’t by industry. It’s by size.

Here’s the number that should worry a small operator more than any headline about robots.

The U.S. Census Bureau tracks AI use across the economy every two weeks. As of May 2026, 19.8% of American businesses reported using AI in any business function. Among firms with 250 or more employees, it was 37%. Among firms with four or fewer employees, under 20%.

And the trend line is the real story. Between December 2025 and May 2026, AI use rose among firms with at least 20 employees — and did not change significantly among firms with fewer than 20.

Big companies are pulling away. Not because they have better ideas, but because they have someone whose job it is to wire this in. Most small businesses don’t. That’s the whole gap.

Why most of it doesn’t stick

Adoption isn’t the same as results. MIT’s Project NANDA found that roughly 95% of generative-AI pilots produce no measurable return — and critically, the failure mode isn’t the model. It’s workflow integration. The tool works. Nobody built it into how the work actually moves.

The Census Bureau’s worker survey shows the same pattern from the other side. About 55% of U.S. workers have used AI for at least one job task. Among those who used it in the past week, 31% saved one to two hours — but 10% saved no time at all, and another 3% said it cost them time.

That’s what unassisted adoption looks like: some real wins, a lot of wash. Drafting an email in ChatGPT saves you four minutes. It does not shorten your close, route a lead before it goes cold, or stop the same customer detail from being typed into three systems by three people. That takes a workflow, not a subscription.

Where the hours actually are

There’s a second finding worth sitting with. A peer-reviewed NBER study of 5,179 customer support agents found AI raised issues resolved per hour by 14% on average — but 34% for the least experienced workers, and close to zero for the most experienced.

Read that as a staffing answer, not a replacement one. AI closes the gap between your newest hire and your best one. In a market where the labor pool is shrinking and the good candidate costs more than you budgeted, that is the single most useful thing it does.

In practice, those hours come out of four places:

• Operations Documenting how work actually moves, cutting the handoffs that cause rework, and putting scheduling, vendor coordination, and capacity checks on rails.
• Accounting & Finance Automated invoicing and AR follow-up, a month-end close that doesn’t eat a week, and cash-flow reporting you can read on your phone.
• Marketing Content pipelines that keep running when the week gets busy, and lead routing that gets an inquiry to a human before it goes cold.
• Customer Service Intake that captures the right details the first time, AI-drafted responses your team reviews and sends, and triage that flags the accounts worth a personal call.

Hire full-time leaders for all four and you’re past $400,000 in salary before benefits. Fractional means you get the senior judgment and the systems that come with it, for the fraction of the week your business actually needs.

Why local counts for more than it used to

There is no shortage of firms that will sell you an AI platform. There is a real shortage of people who will sit at your counter on a Tuesday and watch how the work moves before recommending anything.

That difference is exactly the 95% failure mode. Most automation projects are designed for a business that doesn’t exist yet — a tidy, well-documented, fully-staffed version of yours. A partner who’s been in the room knows about the spreadsheet your bookkeeper maintains on the side, the customer who only calls the landline, and the two steps everybody skips in August.

Our business only works if yours does

We don’t sell software and disappear, and we don’t build things only we can maintain. The 95% number isn’t an argument against automation — it’s an argument against automation without someone accountable for whether it lands.

So the measure isn’t whether you’re “using AI.” It’s whether the close got faster, the intake queue got shorter, and your Thursday got quieter. If those don’t move, nothing else we did counts.

A note on the numbers above: adoption surveys disagree because they ask different questions — anything from “have you ever opened ChatGPT” to “does AI touch your production process.” Every figure here links to the primary source so you can check the definition yourself.