Every few weeks a client forwards us an article about a company that replaced a whole department with AI. Then comes the question. “Are we behind?” Fair question. It deserves a better answer than a vendor demo.
Here is the honest version. AI use in American business is real and growing, and it looks almost nothing like the highlight reel. Most of it is writing, summarizing, and answering customers faster. Most of it lives in one or two departments, not across the company. And the numbers move fast enough that any figure you read today is a snapshot, not a permanent fact about your industry. So let’s start with the patterns that hold up, then the data, then the question that actually matters.
The Patterns That Stay True Even When the Numbers Change
Adoption statistics have a short shelf life. The shape of adoption does not. Remember these five things.
- Adoption is uneven by industry. Information companies, financial firms, and professional services outfits sit far ahead of retail, construction, and transportation. A national average tells you almost nothing about your competitive position.
- Bigger companies adopt faster. A firm with a few hundred employees usually has someone whose job includes evaluating new tools. A twelve person shop does not. That gap is a staffing story, not a technology story.
- The actual work is language work. Drafting, summarizing, searching internal documents, cleaning up customer communication. Not robots. Not machines making decisions on their own.
- Even adopters use it narrowly. Companies that answer yes to “do you use AI” are typically using it in a handful of places, often marketing and IT, and nowhere else.
- Expectation runs ahead of reality. In survey after survey, the share expecting to use AI soon exceeds the share using it now. Intent is not deployment.
What the Federal Numbers Actually Said, and When
The most useful source here is not a vendor report. It is the U.S. Census Bureau’s Business Trends and Outlook Survey, which asks a very large sample of American businesses whether they used AI to help produce goods or services in the prior two weeks. It runs continuously and breaks results out by industry and company size.
In an analysis published by the U.S. Census Bureau in May 2026, overall AI usage among American businesses hovered between 17 percent and 20 percent across collection periods running from December 14, 2025 through May 3, 2026, while 20 percent to 23 percent expected to be using it within the following six months. That same 2026 Census Bureau analysis reported 37 percent usage among firms with 250 or more employees and less than 20 percent among firms with four or fewer.
The industry spread in that 2026 Census Bureau analysis is the part worth writing down. The Information sector reported 39.7 percent current use, Finance and Insurance 33.9 percent, and Retail Trade roughly 14 percent. That is close to a threefold gap at the same moment. If you run a retail operation and feel behind, you may just be reading somebody else’s numbers.
A 2026 U.S. Census Bureau working paper on AI diffusion, CES-WP-26-25, confirms how narrow the usage is. Among firms using AI, the most common functions were sales and marketing at 52 percent, strategy and business development at 45 percent, and IT at 41 percent. That same 2026 paper found 57 percent of users applied AI in three or fewer business functions and 65 percent limited it to three or fewer tasks, and it identified writing, document analysis, and information search as the leading generative AI uses at the task level.
Read those figures as a dated snapshot, because that is what they are. The percentages will have moved by the time you read this. The shape almost certainly has not.
Why the Loud Examples Are Not Representative
The stories that travel are the dramatic ones. That is how attention works, but it means the examples reaching your inbox are a badly skewed sample.
- They started from a different place. A software company with forty engineers and clean data is not a comparison for a distributor running an industry specific system from 2011. Same technology, different starting line.
- Nobody writes up the projects that fizzled. Quiet abandonment does not make a press release. You see the winners and none of the attempts that went nowhere, so the success rate looks far higher than it is.
- The word “AI” covers an enormous range. One headline means a chatbot answering routine questions. Another means forecasting math that has existed for decades and got rebranded. Ask what the tool actually does before comparing yourself to it.
How to Find Out What Your Actual Peers Are Doing
National averages are interesting. What your competitors down the road are doing is useful. Here is how to get the second kind for free.
- Ask your trade association. Most run member surveys and conference sessions. That data is narrower than a federal survey and far more relevant, because everyone in it does what you do.
- Join a peer group or owner roundtable. Owners tell each other things a press release never captures, including what they quietly turned off six weeks later.
- Ask your software vendors a specific question. Not “do you have AI features,” which always gets a yes. Ask what their other customers in your industry use those features for, and how many turned them on. The answer, or the dodge, tells you a lot.
- Ask your own team. People are usually already using these tools, just not through anything you approved. Worth knowing, and also a governance issue we cover in our post on shadow IT.
- Use local business organizations. Chambers of commerce and economic development groups around Denton County run enough owner-level programming that you can usually find someone in your sector who already tried it.
The Better Question Is Which Task Goes First
Here is the contrarian part. Your industry’s adoption percentage is fine trivia and a poor basis for a decision. Nobody ever improved a business by matching an average. The question that produces a result is smaller: which specific task in your business is worth handing over first?
A good first candidate has most of these traits.
- It is mostly writing or reading. That is where these tools are strongest, and it matches what the federal data shows.
- It happens over and over. A weekly task gives you enough repetitions to judge it. A quarterly one will not.
- A person still reviews the output, and a mistake is recoverable. First drafts, not final answers. Start where an error costs a rewrite, not a client.
- You can measure it. Hours, turnaround time, backlog. Pick the number before you start, so you are not arguing about feelings in ninety days.
Pick one. Give it a real trial with a couple of people. Decide honestly whether it earned its keep. Then pick the next one. That works regardless of your sector’s adoption rate, and it is the approach we lay out in our guide to preparing your team for AI without the hype.
The Bottom Line
AI adoption is uneven, narrow, and concentrated in language tasks. Your industry probably sits well away from the national average, and the companies in the headlines are not running your business. That does not mean ignore the technology. It means stop measuring yourself against a number that expires next quarter and start measuring one task you actually handed over. One task, done deliberately, beats a strategy deck.
If you want help picking the right first task, and doing it without creating a security or data mess, we are glad to walk through it. We work with small and mid-sized businesses across Denton County, and we would rather talk you out of a bad fit than sell one. Contact us today.
Sources:
Comments are closed