Small business AI adoption stalled. What that means for you
The Census Bureau puts US business AI use at 19.8%, with firms of four or fewer employees under 20% and flat for six months. Meanwhile voice model pricing fell and quality rose. That gap is a distribution problem, and it is the whole reason an AI agency has a market.
Lumina8 min read
About one in five US businesses used AI in the two weeks ending May 3, 2026. The Census Bureau puts the national rate at 19.8%, and the interesting part is not the headline number. It is the split underneath it. Firms with at least 250 employees are at 37%. Firms with four or fewer employees are under 20%, and that number did not move in a statistically meaningful way between December 2025 and May 2026.
Six months of flat adoption among the smallest firms, while the tools got cheaper and better every few weeks. If you sell AI systems to local service businesses for a living, that gap is the single most important number in your market, and it does not mean what most agency owners assume it means.
What the Census data actually says
The Business Trends and Outlook Survey is not a vendor report. The Census Bureau runs it against roughly 1.2 million businesses on a two-week cycle, representative of all US employer businesses except farms. When a survey that size says the small-firm line is flat, the small-firm line is flat.
Here is what the May 2026 analysis reported, covering data collected from December 14, 2025 through May 3, 2026:
The national rate hovered between 17% and 20% for the whole six months.
37% of firms with at least 250 employees reported using AI. 32% of firms with 100 to 249 employees did.
Less than 20% of firms with four or fewer employees reported using AI.
Use rose among firms with at least 20 employees. It did not change significantly among firms with fewer than 20.
Retail trade sits around 14%, below the national average. The Information sector is at 39.7% and Finance and Insurance at 33.9%.
The companion Census working paper, CES-26-25, adds the detail I keep coming back to. Among firms that do use AI, 57% use it in three or fewer business functions. The most common are sales and marketing at 52%, strategy and business development at 45%, and IT at 41%. Adoption, where it exists, is narrow.
Weight the same data by employment and the firm-level 18% becomes 32%. That is a polite way of saying the average American worker is far more likely to touch AI at work than the average American business is to have adopted it. Big employers pull the weighted number up. The corner plumbing company does not.
Why a flat line is not a dead market
The reflex reading is pessimistic. Small businesses tried AI, it did not work, the market is not there.
I do not buy it, and the reason is on the supply side. Look at what shipped in the same window the small-firm line stayed flat.
OpenAI made GPT-Live 1 generally available on September 10, 2026, priced at $0.05 per minute for voice sessions, billed per second. In July it shipped GPT-Realtime-2.1 with better alphanumeric recognition and interruption handling, which is the difference between an agent that can take a phone number and one that cannot. On July 30 it cut GPT-5.6 Luna by 80%. On August 21 it cut GPT-5.6 Sol input pricing by 20% and output by 33%. We wrote about what full-duplex voice changes for agencies when GPT-Live first landed.
So in six months the cost of running a voice agent fell, the quality rose, and adoption among four-person businesses did not budge. Cost was not the blocker. Capability was not the blocker.
The blocker is that nobody showed up.
A roofing company owner is not reading API changelogs. He is not going to evaluate speech-to-speech models against his missed-call rate. He is in a truck. The gap between 37% at large firms and under 20% at four-person firms is mostly a distribution gap, and distribution is a job someone has to do in person, on the phone, with a demo in hand.
That is the business. It is also why the flat line is good news for you and bad news for anyone waiting for the market to educate itself.
What the numbers change about how you sell
Three practical consequences.
Your prospect is not evaluating vendors. At under 20% adoption, most of the businesses you call have no AI system to replace. You are not competing against a competitor's agent. You are competing against the answering machine and the receptionist who leaves at five. Stop writing outbound that assumes a comparison. Write outbound that assumes ignorance of the category, which is the honest starting point for four out of five prospects.
Narrow beats broad. The working paper found 57% of adopting firms use AI in three or fewer functions, and the top function is sales and marketing. Firms that adopt do it in one place first. So sell one place. Missed calls. After-hours booking. Nothing else in the first conversation. A proposal listing eight AI capabilities reads as a research project to someone who has adopted zero.
Show, do not explain. The reason large firms adopt faster is that they have someone whose job is to evaluate this. Your prospect has nobody. The only compression available is putting a working thing in their hands during the call. When a prospect can call the agent themselves and hear it answer in the language of their own trade, you skip the entire evaluation step they have no capacity for. We built the vertical packs for exactly this reason, and it is the mechanism behind our HVAC playbook too.
The part of this that argues against you
Now the honest section, because a flat adoption line cuts both ways and I would rather you hear it from us.
A market that has not adopted is a market that has not budgeted. There is no line item to redirect. Every deal you close creates a new expense in a business that was not planning one, which is a harder sale than replacing a tool they already pay for. Longer cycles, more no-decisions, more people who say yes and then go quiet.
The survey also has real limits. The question changed in November 2025, from AI use "in producing goods or services" to AI use "in any business function," so year-over-year comparisons across that boundary are shaky. It is self-reported. A plumber who pastes job descriptions into ChatGPT on his phone may well answer no, because he does not think of that as the business using AI. Real informal use is almost certainly higher than 19.8%, and informal use is not the same as a buying signal.
And there is a reading of the flat line I cannot rule out. Maybe four-person businesses are flat because the honest value at that size is small. A company doing forty calls a week may not lose enough to missed calls to justify a monthly fee. If you sell into the very bottom of the market, you will hit that wall, and you should qualify on call volume before you qualify on interest. We say who this is for on our own site and it is not everyone.
One more. "The market is underpenetrated" is the oldest reason to start a company and a bad reason on its own. Underpenetrated markets are sometimes just hard. The Census data tells you the opportunity exists. It tells you nothing about whether you can execute on it, and the execution is cold outbound, demos, and follow-up. Our post on getting your first ten clients is the unglamorous version of that.
How to use this in a sales conversation
Do not quote the statistic at your prospect. Nobody buys because of a Census table.
Use it to set your own expectations. Assume the person on the phone has never seen a working AI receptionist. Assume they have heard the phrase and formed a vague opinion from a podcast. Assume no internal champion, no evaluation process, no budget line. Then build the call around one demonstrated outcome in their trade, with a number they can check against their own book.
The adoption gap is not a talking point. It is a description of who answers the phone.
FAQ
How many small businesses actually use AI right now?
The Census Bureau's Business Trends and Outlook Survey put the national rate at 19.8% for the period ending May 3, 2026, with firms of four or fewer employees under 20%. That figure covers formal business use reported by the firm. Informal use by owners and staff is likely higher and the survey does not capture it well.
Is small business AI adoption growing?
Not at the small end. Census found use rose among firms with at least 20 employees between December 2025 and May 2026, and did not change significantly among firms below that threshold. Growth is concentrated in larger firms and in the Information and Finance sectors.
Does low adoption mean AI does not work for small businesses?
The data does not say that. It measures adoption, not outcomes. The same six months saw voice model pricing fall and quality rise, which argues that the constraint is awareness and distribution rather than capability. That said, a business with low call volume may genuinely not clear the bar, and you should check volume before you pitch.
Which functions do small businesses adopt first?
Sales and marketing, by a wide margin. The Census working paper found 52% of adopting firms use AI in sales and marketing, and 57% of adopters use it in three or fewer functions at all. Lead with one function, not a platform.
Should I wait for the market to warm up before starting an agency?
That is a bet that someone else does the education and leaves the margin for you. The evidence so far is that the small-firm line stays flat until a person shows up with a demo. If you are waiting, you are waiting on yourself.
Where to start
Open the live demo on our homepage, switch the trade to whichever one you call most, and talk to the agent for two minutes. That is the exact experience your prospect gets, minus your logo on it. If it holds up against the objections you actually hear on the phone, start your workspace and build one branded to a real prospect before you make your next round of calls.
