What Small Businesses Get Wrong About 'AI Strategy'

Everyone has an AI story now. The board asked about it. A competitor mentioned it in a press release. A vendor added “AI-powered” to the name of a feature you’ve had for three years. So you do what a responsible owner does: you start working on an AI strategy. And that, ironically, is often where things go wrong.

The Statistic That Should Worry Everyone

McKinsey published their annual State of AI report in 2025, and the headline number got a lot of attention: 88% of large organizations now use AI in at least one business function, and 78% use generative AI — up from 65% in early 2024 and just 33% in 2023 (McKinsey, “The State of AI,” 2025).

The number that doesn’t get as much coverage is this one: only about 5–7% have actually scaled AI company-wide. And only about a third have scaled beyond early pilots.

So 88% started. 7% finished. The other 81% are somewhere in between — running experiments, paying for platforms they’re not fully using, reporting to their boards that they have an AI initiative underway. That gap between adoption and execution is the most important data point in that report, and almost nobody is talking about it.

What an “AI Strategy” Usually Looks Like in Practice

I’ve seen this pattern enough times to describe it precisely. A business owner or executive decides the company needs an AI strategy. Someone is assigned to evaluate options. They come back with a list of tools. The company picks two or three, signs contracts, and announces internally that they’re now using AI. Six months later, a handful of employees are using the tools occasionally. The metrics haven’t changed. The initiative quietly loses momentum.

What happened? Nothing went wrong exactly. The tools are probably fine. The employees aren’t resisting. The problem is that the entire effort started with “what AI should we buy?” instead of “what specific business problem do we need to solve?” Those are not the same question, and they lead to completely different outcomes.

Buying AI features without a specific problem to attach them to is like buying a crane because construction companies are using them. Maybe you need a crane. But if you don’t have a job that requires one, you’ve just bought an expensive thing that sits in your parking lot and makes visitors ask questions.

The Core Mistake: Treating AI as a Product Instead of a Method

Here is the mindset shift that separates the 7% who actually scale from the 81% who stall. The 7% don’t think about AI as something they buy. They think about it as something they use to solve a problem. The goal is never to “implement AI.” The goal is to fix a broken quoting process, reduce the time from order to invoice, give their service team visibility into customer history without switching between four screens. AI may be part of how they get there, but it’s the method, not the outcome.

Most small businesses fall into the trap of buying AI features because that’s how AI is marketed. Every SaaS vendor in 2026 has “AI” somewhere on their pricing page. The AI features are often real, often useful in isolation, and almost never the thing that actually changes how your business operates day to day.

The Federal Reserve noted in their 2026 FEDS Notes on AI adoption that by mid-2025, small businesses were adopting AI faster than large firms while large-firm adoption plateaued (Federal Reserve, FEDS Notes, 2026). That’s an encouraging trend. But adoption rate isn’t the same as results rate. You can adopt AI and still not change anything meaningful about how your business runs.

The Three Places Small Business AI Efforts Stall

Based on what I see with SMB clients across industries, most AI efforts stall in one of three places.

• Stall point one: No owner. AI initiatives without a specific person accountable for a specific outcome die in committee. Someone needs to own the problem being solved, not the technology being evaluated. “AI champion” is not a job. “Fix the customer onboarding process” is.
• Stall point two: Too broad, too fast. The company tries to roll out AI across three departments simultaneously instead of proving value in one workflow first. Nothing gets deep enough to matter. Everyone has a surface-level experience and walks away thinking the technology isn’t ready.
• Stall point three: Rented tools, generic outputs. The business buys AI features embedded in existing SaaS platforms. Those features are configured for the average company, not for them. The output doesn’t fit their terminology, their process, or their data structure. Employees stop using it because it’s more work to adapt the AI’s output than to just do the thing themselves.

Why Off-the-Shelf AI Features Almost Never Stick

The SBA found that among small businesses not using AI, about 77% say they see no applicable use case for their business. That number surprised a lot of people. I wasn’t surprised. I’ve talked to too many operators who have looked at the AI features in their existing tools and genuinely could not see how any of it would help them run their business better. And in many cases, they’re right.

Generic AI features are built for generic use cases. Your business isn’t generic. Your quoting process has quirks that reflect fifteen years of hard lessons. Your customer data is structured the way it is for reasons. Your team has workflows that evolved because the standard way of doing things didn’t work for your product mix, your geography, your sales cycle. An AI feature that was designed for the broadest possible market is going to miss all of that.

Contrast that with software built specifically for your business. The AI assistance can be trained on your terminology, connected to your actual data, and calibrated to produce outputs that your team can use without translation. That’s not a subtle difference. It’s the difference between a tool your team uses every day and one they ignore.

What to Do Instead

The alternative to buying an AI strategy is building a problem list and working through it. Start by identifying the three workflows in your business where the gap between “how it currently works” and “how it should work” costs you the most. Measure the cost in time, error rate, customer complaints, or employee frustration — whatever unit makes the problem concrete. Pick the worst one.

Then ask: what would have to be true for this workflow to work the way it should? What data does it need? What decisions does it make? What does the output look like? Write that down. That document is more valuable than any AI strategy deck, because it describes an actual outcome rather than a technology preference.

From there, build something that solves exactly that problem. Not a platform. Not a suite. One tool, owned by your company, connected to your data, calibrated for your process. Measure the result. Then move to the next problem on your list. That’s not a less ambitious approach than building an AI strategy. It’s a more disciplined one. It’s also the approach that actually produces results instead of PowerPoint slides.

The Businesses Getting This Right

The small and mid-sized businesses that are genuinely ahead in 2026 share a few traits. They don’t use the word AI in internal conversations about their software. They talk about the problem they’re solving. They have specific owners for specific outcomes. They measure results in business terms, not technology terms. And they build software they own rather than subscribing to features that may or may not be there next year. The McKinsey number — 5–7% scaled — isn’t because the technology is hard. It’s because most organizations start with the wrong question. Start with the right one, and that number is beatable.

Sources

• McKinsey & Company — The State of AI (2025)
• Federal Reserve FEDS Notes — Monitoring AI Adoption in the U.S. Economy (2026)
• JPMorgan Chase Institute — Understanding AI Use by Small Businesses (2025)