A Census Bureau survey of US employer businesses, excluding farms, asked businesses using AI in any business function what they changed in order to use it. Those with 1 to 4 employees were less likely than those with 250 or more to report training current staff. On that same question, 65.5% of AI-using businesses with 1 to 4 employees selected the answer for making no changes, against 48.5% of those with 250 or more.
Adopting a tool and changing how work gets done are two different things. In a supplement to its Business Trends and Outlook Survey, collected from November 17, 2025 to February 8, 2026, the Census Bureau asked employer businesses that used AI in any business function what they had changed in order to use it. Figures below come from the Census table “Employment Response Estimates” and its standard errors, downloaded October 3, 2026. Census describes the survey as “representative of all employer businesses in the U.S. economy, excluding farms”.
Census has already published AI use by company size and reported that larger companies use AI more. In this supplement, 17.6% of employer businesses with 1 to 4 employees reported using AI in any of its business functions in the last two weeks, against 30.6% of those with 250 or more employees. This study looks at the changes AI-using businesses reported making in order to use it.
Training current staff: an 18-point gap
Among businesses that used AI in any business function, 13.4% of those with 1 to 4 employees said they trained current staff to use AI. Among businesses that used AI in any business function and had 250 or more employees, 31.4% did. The smaller group was lower by 18.0 percentage points (approximate 95% interval: 15.9 to 20.1 points lower). This was a pre-registered comparison, and the interval excludes zero. The item asks about training current staff. The table does not show whether that item means the same thing in a business with 1 to 4 employees as in a business with 250 or more. The 18.0 point gap is a difference in answers, not a measured difference in training effort. Size classes differ in industry and occupation mix. Differences by size reflect both. Nothing here isolates an effect of size.
On the same question, respondents could select more than one change, so these shares are not parts of one whole and can add to more than 100%. Every answer Census published is shown below, so no answer was picked out after the results were seen. Apart from training current staff, these figures are descriptive.
Changes made in order to use AI, among businesses that used AI in any business function (source: U.S. Census Bureau, BTOS AI supplement 2025 to 2026, sheet “Employment Response Estimates”, downloaded October 3, 2026. Answer wording is Census’s.):
| Answer | 1 to 4 employees | 250 or more employees |
|---|---|---|
| Trained current staff to use AI | 13.4% | 31.4% |
| Hired staff trained in AI | 1.0% | 6.8% |
| Purchased computing power or specialized equipment or software | 7.6% | 16.1% |
| Purchased cloud services or cloud storage | 8.5% | 12.7% |
| Changed data collection or data management practices | 6.7% | 11.3% |
| Developed new workflows | 15.0% | 21.1% |
| Used vendors or consulting services to install or integrate AI | 3.1% | 10.3% |
| Other | 5.1% | 4.4% |
| This business did not make any changes to use AI | 65.5% | 48.5% |
Cited lack of knowledge: no clear difference between the smallest and largest
The study’s second pre-registered comparison asked about businesses not planning to use AI in the next six months. Among businesses not planning to use AI in the next six months, 21.2% of those with 1 to 4 employees cited a lack of knowledge about AI’s capabilities, against 20.0% of those with 250 or more employees. The difference, 1.2 percentage points, has an approximate 95% interval of negative 1.3 to positive 3.7 points, which includes zero. The pre-registered comparison does not show a clear difference between businesses with 1 to 4 employees and businesses with 250 or more on this reason. The size classes in between are outside that comparison. Among businesses not planning to use AI in the next six months, the middle size classes’ point estimates for citing lack of knowledge about AI’s capabilities run from 23.0% to 25.0%, which is descriptive only. This result does not show that the share is the same at every size.
On the reasons question, respondents could select more than one reason, so these shares are not parts of one whole. Every answer is shown below. Apart from lack of knowledge, these figures are descriptive. The most common answer in both size classes was that AI is not applicable to the business.
Reasons for not planning to use AI, among businesses not planning to use AI in the next six months (source: U.S. Census Bureau, BTOS AI supplement 2025 to 2026, sheet “Employment Response Estimates”, downloaded October 3, 2026. Answer wording is Census’s.):
| Answer | 1 to 4 employees | 250 or more employees |
|---|---|---|
| Too expensive | 6.7% | 9.0% |
| AI is not a mature enough technology yet | 12.6% | 19.4% |
| Lack of knowledge on the capabilities of AI | 21.2% | 20.0% |
| Concerns about privacy/security | 20.1% | 29.7% |
| Concerns about bias | 8.9% | 9.1% |
| Lack of skilled workforce | 6.4% | 8.1% |
| Lack of required data | 5.0% | 4.5% |
| Laws and regulations prevent or restrict use of AI | 2.8% | 4.9% |
| Previous or current use of AI did not meet expectations | 3.3% | 2.3% |
| AI is not applicable to this business | 63.3% | 42.2% |
| Other | 7.6% | 14.6% |
Employment
Census asked businesses that used AI in any business function how, in the last six months, the use of AI affected the business’s total employment. Among businesses that used AI in any business function, 95.7% of those with 1 to 4 employees and 94.8% of those with 250 or more answered no change. This is the businesses’ own report, not a measured change in jobs.
Free 20-Minute Operations Review
Dealing with a specific operational bottleneck? Kamyar Shah works with founders and CEOs to identify the root cause and build a fix.
What the numbers mean for a small business using AI
This section and everything after it, up to Method, are interpretation and guidance from operating practice. They are not findings of this study.
Among businesses with 1 to 4 employees that used AI in any business function, 65.5% said they made no changes in order to use it, and 15.0% said they developed new workflows. In practice, that pattern often means AI is being used as a tool on top of existing work rather than as a reason to change how the work is done. The data cannot show whether that use is productive. It does show that most of the smallest AI users did not report adjusting anything around it.
The training gap reads the same way. Only 13.4% of AI-using businesses with 1 to 4 employees reported training current staff, against 31.4% of AI-using businesses with 250 or more employees. In a business of one to four people, the person using AI is often the owner, so “training” can look like nothing at all. The practical question is not whether a formal course happened. It is whether anyone has written down how the tool is used, so the use is repeatable when someone else does the work.
Among businesses with 1 to 4 employees that were not planning to use AI in the next six months, 63.3% said it is not applicable to their business, and 21.2% cited a lack of knowledge about AI’s capabilities. Those two answers are not exclusive, and the second is worth noticing. A business that has not looked closely at what the tools do is in a weak position to decide they do not apply.
How to handle it: treat AI as a process change, not a purchase
The steps below are an operating recommendation. This study did not test them, and Census does not endorse them.
1. Pick one workflow, not the whole business. Choose a task that repeats every week, such as drafting quotes, answering routine customer questions, or reconciling a report. Write down how it is done today, step by step, and how long it takes.
2. Change the workflow on purpose. Decide which steps the tool takes over, which steps a person still checks, and what the finished output must look like. That written version is the new standard procedure. Without it, each use of the tool is less likely to leave a repeatable process behind.
3. Train whoever does the work, including the owner. Training in a small business can be a one-page procedure and a short walkthrough. The test is simple: someone other than the person who set it up can follow the procedure and get the same quality of result.
4. Measure before and after. Time per task, error rate, or turnaround are enough. If the numbers do not move after a month, the tool is not yet changing the work, and the procedure needs another look.
5. For businesses that decided AI does not apply, test that conclusion once. List the tasks that consume the most hours each week and check each against what current tools can do. The conclusion may hold. It is still a decision worth making with the task list in hand rather than by default.
What this data does not tell you
- It does not show that using AI, changing workflows, or training staff improves results.
- Answers are the businesses’ own reports, collected from November 17, 2025 to February 8, 2026.
- The comparisons cover businesses with 1 to 4 and 250 or more employees. The size classes between them are shown only where stated.
- Respondents could choose more than one answer, so shares do not add to 100%.
Using AI and changing the work around it are different decisions. Among AI-using businesses with 1 to 4 employees, 65.5% reported making no changes in order to use AI. As operating guidance rather than a finding: the change in the work, written down and repeatable, is where an owner should expect the value to come from. The survey does not measure whether it does.
Method
Already published by the agency: Census has published AI use by company size and reported that larger companies use AI more. New here: two pre-registered comparisons by company size, one among businesses using AI and one among businesses not planning to use it, with descriptive figures on changes made and reasons given.
Source. U.S. Census Bureau, Business Trends and Outlook Survey, AI supplement 2025 to 2026, file AI_Supplement_Table_2026.xlsx released June 18, 2026 (Census tip sheet CB26-TPS.38), sheets “Employment Response Estimates” and “Employment Standard Errors”, downloaded October 3, 2026. Size classes are 1 to 4, 5 to 9, 10 to 19, 20 to 49, 50 to 99, 100 to 249, and 250 or more employees. The comparisons use the smallest and largest classes. The middle classes appear once, as a descriptive range.
Who answered which question. Use of AI in the last two weeks was asked of all employer businesses. Changes made and employment effects were asked of businesses that used AI in any business function. Reasons for not planning to use AI were asked of businesses not planning to use it in the next six months.
Statistics. Each difference uses the two published standard errors. Census does not publish the covariance needed for these contrasts, so the approximate intervals assume zero covariance between size-class estimates, with z equal to the difference divided by the square root of the sum of the squared standard errors. Intervals are approximate 95% intervals. No multiplicity adjustment was made. Answers Census marked as suppressed were not used.
Protocol. The analysis plan, including the two primary comparisons, was fixed in a dated protocol before any size-class estimate was computed. Later amendments are listed in it: https://kamyarshah.com/public-data-protocol/
The agencies did not review or endorse this analysis. Kamyar Shah sells fractional COO and CMO services.
Questions about this analysis can be sent through https://kamyarshah.com/contact/.
