Nearly half of small employer firms use AI. Few have fully integrated it.
Nearly half of small employer firms use AI. That’s the headline from the Federal Reserve’s 2026 Small Business Credit Survey report: 46%, to be exact.
It sounds like a revolution. Then you read the next chart.
Among those AI users, 49% are still experimenting. Another 44% have integrated AI into some business processes. Just 7% have fully integrated it into the business.
Plenty of businesses have started. Few have finished.
I’ve spent more than 25 years building and running businesses. I’ve also automated my whole job with AI, not just parts of it. So the gap between using AI and working differently because of it is hard for me to ignore.
What the Fed’s three levels mean
The survey asked AI-using firms which level fits them. The report gives these short descriptions.
Experimenting (49%): testing tools or running pilots.
Partially integrated (44%): AI is used in some business processes.
Fully integrated (7%): AI is embedded in business functions.
These are each firm’s own answers, not an outside audit. The full wording is in the survey questionnaire.
Using AI isn’t the same as working differently
First, a caveat. The numbers depend on who’s asking.
Why the Fed and Census numbers differ
Federal Reserve (46%): asks employer firms with fewer than 500 employees whether the business or its employees use AI for work.
Census Bureau (19.8%): asks about AI in business functions, and its sample includes the very smallest firms. Its latest estimate is from early May 2026.
Different questions. Different samples. My read: they’re measuring different things, not contradicting each other. (That’s an interpretation, not a finding.)
Here’s where they point the same way. Census researchers found that 57% of AI-using firms apply it in three or fewer business functions. The most common are sales and marketing (52%), strategy and business development (45%), and IT (41%).
A gym membership isn’t fitness. It’s a membership. Plenty of businesses have the AI version of that.
What the evidence shows when AI is used well
So what happens when AI gets used for real work? Researchers have measured it, mostly in large organizations, which matters, and I’ll come back to that. The last three rows are different: they show where small firms use AI most, not what it delivered.
| Activity | What was measured | Who was studied |
|---|---|---|
| Customer support | 15% more issues resolved per hour with AI assistance (Brynjolfsson, Li & Raymond) | 5,172 agents at one Fortune 500 software company |
| About two fewer hours a week on email among workers who used the tool (Dillon et al.) | 7,137 workers across 66 large firms | |
| Accounting | Monthly close 7.5 days shorter among AI users, an association rather than proof of cause (Choi & Xie) | 79 small and mid-sized firms, 277 accountants |
| Online retail | Sales effects from none detected to 16.3%, depending on the application (Fang et al.) | One cross-border retail platform |
| Sales and marketing | Most common function for AI use: 52% of AI-using firms (usage, not results; Census) | AI-using U.S. firms, Census Bureau |
| Strategy and business development | 45% of AI-using firms (usage, not results) | AI-using U.S. firms, Census Bureau |
| IT | 41% of AI-using firms (usage, not results) | AI-using U.S. firms, Census Bureau |
Each result describes a specific task under specific conditions. Some come with caveats. The email study is a working paper, and three of its four authors worked at Microsoft, which makes the tool studied. The accounting data came through an AI accounting software provider. And most of these studies involve large organizations. The Census rows show adoption, not outcomes.
None of this is a forecast for your business. It does show something important, though. AI can improve real work.
Where AI helps, and where it quietly doesn’t
The most useful study here is the one with a failure in it.
Researchers gave 758 consultants at Boston Consulting Group a set of realistic tasks. The 18 tasks inside AI’s strengths covered:
- Creative idea generation
- Analytical thinking
- Writing
- Persuasion
On those, consultants using AI completed 12.2% more tasks and finished 25.1% faster. Quality improved too.
The researchers also included one task built to sit outside those strengths: a business problem that required combining spreadsheet data with interview evidence. They chose it on purpose, so the result shows that a boundary exists, not how often AI falls short in everyday work. On that task, AI users were 19% less likely to reach the correct answer.
Same tool. Different task. Opposite result.
That’s the part I’d put on a few conference slides. Being good at a task doesn’t make AI good at the whole job.
Example (an illustration, not a study)
Take a real estate development firm. The research supports AI helping with writing and analysis tasks. Whether AI can profitably run that firm’s site selection, entitlement tracking, lender reporting, and contractor coordination is a separate question. It needs testing in that business, with those projects.
Doing a task better isn’t the same as running a better business.
A tool can improve a task. Whether it improves a business depends on how the work around it changes: who does what, where information goes, and how results get checked. That’s my take, and the research doesn’t settle it.
Most of this research comes from large organizations. That the gains transfer to a company with 25 employees is a reasonable hypothesis. It isn’t established.
How I automated my job
I’ve spent more than 25 years in software, across CRM, BI, cloud platforms, low code, and AI. When generative AI took off in November 2022, I started applying it to software sales, marketing, product, and product-led growth.
I tested every CRM and product-led growth use case I could think of and ran several pilots. In 2023, I worked on generative AI solutions for about 20 startups. Along the way, my sales and sales engineering role became an exercise in automating myself out of a job. It wasn’t the first time I’d been hired to do that.
This is my own experience, not a study. It shows where I’m coming from, not what you should expect.
Where to start
If you run a small business, the research still points somewhere practical.
My read of the evidence: it’s strongest for customer support, document work, finding internal information, and marketing. Those are good places to look first. Sales and marketing, strategy, and IT are where small firms use AI most, but common use isn’t proof it pays off. None of these is guaranteed to work for you.
Start with one real problem. Not “we should use more AI.” Something specific. Replies take two days. Proposals take a week. The same question gets answered fifteen times.
Then measure it. Before and after. And count the whole cost: setup, review, training, and somebody’s time. An hour saved isn’t automatically an hour of cash.
Started isn’t finished
The Fed’s numbers say most of these businesses have started with AI. The research says it can help. The gap between the two is the work.
The goal isn’t more AI. It’s more value from the people and resources you already have.
If you’d like to talk about where your business sits on this, book a 30-minute call. No pitch. We’ll talk about where things feel slow, costly, or stuck. I work as a fractional AI expert for small business owners, and you can see how it works on my homepage.
Appendix: Sources and claims
All figures were checked against the linked source on November 5, 2026. “Interpretation” marks my own reading, not a finding.
| Claim in the article | Source and status |
|---|---|
| 46% of employer firms (fewer than 500 employees) report that the business or its employees use AI for work; N=5,248 | Federal Reserve Banks, 2026 Report on Employer Firms: Findings from the 2025 Small Business Credit Survey, p. 27 (DOI) |
| Among AI users (N=2,199): 49% experimenting, 44% partially integrated, 7% fully integrated | Same report, p. 27 and executive summary (p. ii). The summary says “about half” for experimenting; 49% is read from the chart. |
| Overall business AI use was 19.8% as of May 3, 2026 | U.S. Census Bureau, Business Trends and Outlook Survey story, May 2026 |
| 57% of AI-using firms use AI in three or fewer functions; top functions are sales and marketing (52%), strategy and business development (45%), IT (41%) | Bonney et al., Census Bureau CES Working Paper 26-25, April 2026 |
| AI assistance raised issues resolved per hour by 15%; 5,172 agents at one Fortune 500 software firm | Brynjolfsson, Li & Raymond, “Generative AI at Work,” Quarterly Journal of Economics 140(2), 2025, DOI; preprint |
| Workers who used the tool spent about two fewer hours a week on email; 7,137 workers, 66 firms | Dillon et al., NBER Working Paper 33795. Not peer-reviewed. Three of four authors were employed by Microsoft at the time. The effect across everyone given access, including non-users, is smaller. |
| Monthly close 7.5 days shorter among AI users; 79 small and mid-sized firms, 277 accountants | Choi & Xie, Stanford GSB working paper. An association. Firm data came through an AI accounting software provider. |
| Sales effects from no detectable impact to 16.3% across seven retail workflows | Fang et al., Generative AI and Firm Productivity: Field Experiments in Online Retail. Working paper; one cross-border retail platform. |
| 758 consultants: +12.2% tasks completed, 25.1% faster, higher quality inside AI’s strengths; 19% less likely to be correct on the one task chosen to sit outside them | Dell’Acqua et al., “Navigating the Jagged Technological Frontier,” Organization Science, 2026, DOI |
| The Fed and Census figures measure different things | Interpretation. The two surveys differ in wording and sample. |
| The evidence is strongest for support, documents, internal information, and marketing | Interpretation. My assessment of the studies above, not a published rating. |
| Gains may transfer to a company with 25 employees | Hypothesis. Not established by the research cited. |
| Real estate development example | Illustration, not a study. |
| 25+ years building and running businesses; automated my whole job with AI | Adam’s own experience. |