The Small Business AI Adoption Gap

For two years the honest answer to "is my competition using AI?" was that nobody could tell you. The surveys were vendor-run, the samples were unpublished, and the numbers moved forty points depending on who was selling what.

That changed on 26 May 2026, when the U.S. Census Bureau published its read on the Business Trends and Outlook Survey. It is a recurring federal survey with a disclosed collection window — 14 December 2025 through 3 May 2026 — and it found something specific enough to be uncomfortable.

Adoption is not spread evenly. It is stratified by headcount, and the line falls at twenty employees.

The finding, in Census's own words

Here is the sentence that matters, quoted exactly:

"AI use increased among firms with at least 20 employees but didn't change significantly among firms with fewer than 20 employees."

Read that twice, because both halves are doing work. Over roughly five months, businesses above the twenty-employee line moved. Businesses below it did not. Not "moved slower" — did not change significantly.

The size breakdown as of 3 May 2026:

Firm size	Current AI use
250 or more employees	37%
100–249 employees	32%
Fewer than 20 employees	Under 20%
4 or fewer employees	Under 20%

Census does not publish a single point estimate for the small classes — it reports them as under 20%, and we are not going to invent a decimal it did not print.

If you run a plumbing company with nine people, a two-office brokerage, or a lending shop with a handful of originators, you are in the bottom two rows. The competitor who took on a twelfth employee last year is in a different row, and over that five-month window they moved and you did not.

Why the national headline number is the wrong one to quote

You will see "17% to 20% of U.S. businesses use AI" attached to this survey, along with the projection that 20% to 23% expect to be using it within six months. Both figures are real and both come from the same release.

Quoted alone, they are close to useless — and mildly misleading in a direction that happens to be comfortable.

On its own, "one business in five" reads as adoption is low everywhere, there is no rush. The size cut says something almost opposite: adoption is low specifically among businesses like yours, while the firms with staff to spare are already past a third. The national average is an average of two populations moving at different speeds. It describes neither.

A related warning, since you will run into it. The same Census release has been written up under the headline "large firms are the biggest AI users" and, elsewhere, as small firms leading adoption — the second built on a different cut of BTOS reporting single-digit percentages that do not reconcile with the figures above. We are not going to adjudicate that here. We are telling you which document we used, so you can check it: the Census story linked at the top, published 26 May 2026, from the December-to-May collection. Every number in this article is from that page.

Your industry is a second sorting

Headcount is not the only split. Census also broke out sectors, current use against expected use in six months:

Information: 39.7%, rising to about 42%
Finance and insurance: 33.9%, rising to about 39%
Retail trade: about 14%, rising to about 17%

Census notes these sectors exceeded or fell short of the national average without significant shifts since December. That is the part worth sitting with. The distance between the fast sectors and the slow ones held steady across the survey window. It is not closing on its own.

Two of those numbers land directly on people who read this site.

If you originate loans, finance and insurance is your sector and it is running at 33.9%, headed for roughly 39%. That is not a distant-future warning — it is a third of your category already, and by the survey's own projection, closer to two-fifths within six months of the collection close. When a borrower calls three lenders on a Tuesday afternoon, the odds that at least one of them has something answering outside business hours are no longer small. We wrote about what that does to a lending pipeline in more detail, and the sector figure here is the piece that article did not have at the time.

If you run a home-services trade, you are not in a sector Census broke out. Nobody has published your number. What you have is the size finding, and trades skew heavily under twenty employees — which puts you in the row that did not move.

What this data does not say

Now the caveat, because leaving it out would make everything above worth less.

Census measured adoption, not results. The survey asked businesses whether they use AI to produce goods or services. It did not ask whether it worked, what it returned, or whether the firms adopting it are outperforming the ones that are not. A 37% adoption rate among large firms is evidence that large firms are adopting. It is not evidence that they are winning.

Anyone converting these percentages into a revenue-loss figure for your business is doing something the data does not support. That includes us. There is no line in this release that lets you price your own inaction, and we are not going to build a calculation on an input nobody has measured.

What the data does establish is narrower, and still worth having:

The gap is real, federally measured, and defined by headcount.
It widened over the December-to-May window rather than narrowing.
Your sector position is a separate and independently measurable thing.

That is enough to make "we'll look at it eventually" a decision rather than a default. It is not enough to tell you what to buy.

The one number that is actually yours

Here is the exercise, and it costs you an afternoon rather than a subscription.

Pull 90 days of inbound call logs from your phone system. Almost every VoIP platform exports this. You want total inbound, answered, unanswered, and timestamps.

Separate the unknown numbers from the ones already in your CRM. An unrecognised number is disproportionately somebody who has never done business with you. That ratio is your actual exposure, not your headline miss rate.

Count the hours. There are 168 hours in a week. A shop staffed 8 to 5, Monday to Friday, covers 45 of them. The remaining 123 are not a statistic — they are arithmetic, and they are the same for every business your size.

Then ask the question the Census data cannot answer for you. Of the enquiries you reached inside five minutes last quarter, how many became work — against the ones you returned the next morning? If your system cannot produce that comparison, that is your finding, and it matters more than any adoption percentage.

A firm above the twenty-employee line answers more of those 123 hours by throwing bodies at the problem. That is the actual mechanism behind the gap for most of the businesses in this survey. It is a staffing advantage, and it shows up in the data as a technology advantage.

Where an AI Employee fits

Client 4 Life AI is a roster of named AI Employees hired per role. Three of them are relevant to what this data describes.

Ava is a Lead Conversion Coordinator. She answers the call you missed and runs the booking conversation end to end, so the enquiry is qualified and on the calendar while the caller is still deciding.

Kai fields the repetitive procedural questions — hours, process, what happens next, what to bring — that consume a front desk's morning without producing anything.

Tessa holds contact across SMS and email through the days a decision sits open, which is where slow deciders quietly become somebody else's customer.

The limits, stated plainly. This does not fix a business losing work for other reasons — if your real problem is a three-week turnaround, answering faster only means the customer leaves you later in the process. It does not make regulated decisions; it collects and routes, and the boundary is set during configuration. And it does not replace the people you have. Under twenty employees, it is the shift nobody is on.

Pricing is published at $3,000 per employee. It is on the pricing page rather than behind a call, so you can do this arithmetic without talking to us.

Run your own version of this

Take the Census finding, then take your own call log. If your unanswered-unknown-number count over 90 days bothers you, book a call and we will go through your actual data together. If it does not, you have spent an afternoon confirming your phone is covered, which is a real result.

Frequently asked questions

What is the small business AI adoption gap? It is the measured difference in AI use between larger and smaller firms. The U.S. Census Bureau's Business Trends and Outlook Survey, collected 14 December 2025 to 3 May 2026, found that AI use increased among firms with at least 20 employees but did not change significantly among firms with fewer than 20 employees. As of 3 May 2026, 37% of firms with 250 or more employees reported using AI, against under 20% of firms with fewer than 20 employees.

How many US businesses actually use AI? Between 17% and 20% over the survey window, with 20% to 23% expecting to use it within six months. That national figure is an average across firm sizes that are moving at different speeds, so it is not a useful benchmark for any single business on its own — the size and sector breakdowns are.

What percentage of finance and insurance businesses use AI? 33.9% reported current use, with about 39% expecting to be using it within six months, per the same Census release. Finance and insurance is the closest published sector to lending and loan origination, and it runs well above the national rate.

Does the Census data prove AI improves business results? No, and it is worth being clear about that. The survey measured whether businesses use AI, not whether it worked or what it returned. It establishes that a gap exists and is widening by firm size. It does not measure outcomes, and any dollar figure built on top of it is somebody's invention rather than a finding.

Is a small business too small to use AI? Size is what created the gap, not what prevents closing it. The advantage larger firms hold in this data is mostly staffing — more people covering more of the week. For a business under twenty employees the practical question is not whether to adopt AI broadly, but which single uncovered function is costing the most, which for most inbound-driven trades is the phone outside working hours.