Main Street's AI Bet: Small-Business Owners Are Buying In Faster Than They Can Staff or See It

Cover Memo

This report draws on the AI@Work survey (CHIP50 Wave 38.6), a survey of 23,462 U.S. adults (weighted n = 23,485) fielded June 21–July 13, 2026. All percentages are weighted to U.S. Census population targets and are population-representative for the group described; unweighted respondent counts are reported as a reliability guide only.

These are interim estimates. The AI@Work survey launched June 21, 2026 and data collection is ongoing; the figures in this report reflect responses received through July 13, 2026 and will be revised in a later release.

Universe and bases. The report describes small-business owners — respondents who answered yes to owning "a small business that employs non-family" (bus_own). That screener is not available in this data extract, so the owner universe is anchored by the base of the first owner item every owner received (bus_own_ai, unweighted n ≈ 1,241). AI-specific items — how much the business uses AI, budget, planned spending, cost effects, monitoring, and suspected unsanctioned use — were asked only of owners whose business uses or engages with AI (unweighted n ≈ 710–750), and the spending-detail item only of owners who allocated an AI budget (n ≈ 440). Each figure and table states its own base. Given these bases (a few hundred to ~1,240), estimates carry wider uncertainty than the national reports in this series and are not broken out by demographic group.

A framing caution on comparability. This is a population survey of individuals who identify as small-business owners, not an establishment survey of firms. Federal establishment data (the Census Bureau's Business Trends and Outlook Survey, summarized by the SBA Office of Advocacy) put AI use among all small businesses far lower — about 8.8% as of August 2025 (SBA Office of Advocacy). The much higher engagement seen here reflects a different, self-reported, and likely more digitally-connected frame, plus a looser "is AI relevant / do you use it at all" threshold. Read the survey's figures as describing these owners' attitudes and reported behavior, not as an establishment-level adoption rate.

A coding note. The employee-count item (bus_emp) is coded in eight bands (codebook 0 = "0–1 employees" … 7 = "1,000+"); the data return seven of them (codes 1–7 = "2–9" through "1,000+"), with the "0–1" band absent — consistent with the screener, which requires that the business employs non-family. The earlier "offset" ambiguity is resolved by confirming that the data files return raw stored codes for the other 0-indexed owner items (bus_own_ai2, ai_invest), whose own code-0 opt-out categories ("Not sure," "Will not spend") are likewise simply absent from the data. With the mapping settled, this report does report a firm-size breakdown of AI relevance and the skills barrier (Table 9); the direction of those gradients is robust even to a possible one-band shift in the labels. The owner sample skews toward smaller firms (35% at 2–9 employees).

Key Takeaways

Introduction

Artificial intelligence is usually studied through the lens of large enterprises and their workforces, but the overwhelming majority of American employers are small businesses — and how their owners are responding to AI will shape much of the technology's real-world impact. Federal data suggest small firms have been slower to adopt AI than large ones, though the gap is closing: the Census Bureau's Business Trends and Outlook Survey found roughly 8.8% of small businesses (under 250 employees) using AI as of August 2025, up sharply over the prior year and narrowing the distance to large firms (SBA Office of Advocacy).

The AI@Work survey offers a complementary, owner's-eye view. Rather than sampling firms, it asks individuals who own small businesses how they see AI: whether it is relevant to their business, how much they use it, whether they are spending on it and plan to spend more, what it has done to their costs, whether they can find the skills they need, and whether they suspect their own employees are using AI out of sight. The portrait that emerges from this self-selected, self-reporting group of owners is one of enthusiasm outrunning infrastructure. These owners are adopting AI eagerly and intend to spend more — but they are doing so with off-the-shelf tools, unproven savings, a widening skills gap, and limited visibility into how their own staff use the technology. They are, in short, buying in faster than they can staff for it or see it.

AI has arrived on Main Street

Most small-business owners now see AI as relevant to their business — and nearly all of those who do are already using it.

Six in ten small-business owners (60%) say AI is relevant to their business; the remaining 40% say it is not. Among the owners who see it as relevant, adoption is close to universal: 88% say their business already uses AI at least "some," including 38% who use it "a lot," while just 11% use it "not so much" and only 1% "not at all." Put together, roughly half of all surveyed owners run a business that is already using AI in some form, and among those who consider it relevant there are almost no holdouts — the question for this group is no longer whether to use AI but how much.

This is a far higher engagement level than establishment surveys report, and the gap is instructive rather than contradictory. Federal figures count firms and apply a stricter test, yielding single-digit adoption; this survey captures owners who are connected enough to take a national online-recruited survey and asks a looser question about relevance and any use. Both can be true: AI use is still rare across the full census of small businesses, yet common among the engaged, self-identifying owners who respond to surveys like this one — the leading edge of Main Street adoption.

Relevance also climbs steeply with the size of the business. Among the smallest owner-firms — those with roughly a handful of employees (2–9) — 42% say AI is relevant; that rises to 60% among mid-sized firms (10–49) and 77% among owners with 50 or more employees, with the largest firms in the sample the most likely of all (around 84%) to see AI as relevant (Table 9). Whatever establishment statistics say about the very smallest firms, within this owner sample AI engagement tracks headcount: the more people a business employs, the more its owner sees AI as something that matters to it — plausibly because more employees means more processes, roles, and workflows onto which AI can be mapped.

The money is moving — toward tools, not systems

Owners are budgeting for AI and plan to spend more — but the spending is on subscriptions and licenses, not on building their own systems.

Adoption is backed by real, if modest, money. Among owners whose business uses AI, 59% have allocated a dedicated budget or funds for AI (37% have not; 4% are unsure), and the momentum points up: 62% expect their AI spending to increase over the next 12 months — 23% "significantly" and 39% "slightly" — against just 8% who expect to spend less and 30% who expect no change. Intended investment rises directly with how much a business already uses AI, so the heaviest users are also the ones planning the largest increases; this is an expanding commitment, not a one-time experiment.

But the shape of that spending reveals its limits. Asked what their AI investments consist of, owners who budget for AI overwhelmingly name individual subscriptions (70%) and enterprise or team licenses (63%) — the off-the-shelf, per-seat tools any business can buy. Only 25% report investing in custom development, specialized hardware, or dedicated AI staff. Small-business AI, in other words, is overwhelmingly rented rather than built: owners are paying for access to tools like ChatGPT and Copilot, not constructing bespoke systems or hiring AI specialists. That keeps the barrier to entry low, but it also means the capability these businesses are buying is the same capability available to everyone else — and, as the next sections show, it leaves them dependent on skills and oversight they largely do not yet have.

The payoff is still unproven

For all the adoption and spending, AI has not yet delivered clear cost savings to small businesses.

If AI were straightforwardly cutting costs, owners would say so. They do not. Asked how AI has changed their operating costs, small-business owners split almost evenly: 31% say AI has increased their operating costs, 26% say it has decreased them, and 30% report no change, with another 11% saying it is too early to tell and 2% unsure. That as many owners report higher costs as lower ones is a striking counterpoint to the efficiency narrative that usually accompanies business AI. The subscriptions, licenses, and — for some — the staff time to learn and manage these tools are themselves a cost, and for a substantial share of owners those costs have not yet been outweighed by savings.

This does not mean AI is a bad investment for these businesses; the benefits owners perceive may be in quality, speed, or capability rather than in the cost line, and a tenth explicitly say it is too early to judge. But it is a useful corrective to the assumption that small-business AI adoption is primarily a cost-cutting story. On the current evidence, from the owner's own ledger, the cost case is unsettled.

The skills gap widens with use

The more a small business leans on AI, the more its owner feels the shortage of staff who can actually use it.

Owners are candid about a constraint that adoption does not solve: people. Asked to react to the statement that "a lack of staff with AI skills is a barrier" for their business, 42% of owners agree (14% strongly), 29% are neutral, and 29% disagree. On its own that is a plurality naming a skills gap. But the more telling pattern is how agreement scales with use: the weighted mean on the barrier scale climbs steadily from owners whose business uses AI "not at all" to those who use it "a lot" — the businesses most invested in AI are precisely the ones that most feel the shortage of AI-capable staff. The skills gap, in other words, is not a problem that recedes as a business gets more comfortable with AI; it intensifies, because deeper use surfaces more of the work that skilled people are needed to do.

The same is true of firm size. Agreement that a lack of AI-skilled staff is a barrier rises from 26% among the smallest owner-firms (2–9 employees) to 36% among mid-sized firms and 62% among those with 50 or more employees (Table 9) — so the shortage is felt most acutely exactly where it bites hardest, in the larger businesses with more roles to fill. The barrier grows with both how much a business uses AI and how many people it employs.

This connects the owner's dilemma directly to the workforce findings elsewhere in this series: employers report that most workers receive no AI training, workers themselves say they need it and don't know where to get it, and here the owners confirm the demand side of that gap — the businesses adopting AI fastest are the ones that most need AI skills they cannot readily find. For small businesses without the budget to hire specialists or build training programs, that shortage is a real ceiling on what off-the-shelf tools can deliver.

Flying partly blind

Two-thirds of owners suspect their employees use AI without their knowledge — and most rely on trust, not tools, to keep track.

The final piece of the picture is oversight, and here small-business owners describe the same "shadow AI" fog that workers do, seen from the other side of the desk. Among owners whose business uses AI, 67% suspect their employees are using AI without their knowledge — 31% think this happens "frequently" and 37% "occasionally." Only a quarter (25%) are confident their employees do not use AI behind their backs, and just 8% think their employees don't use AI at all. These are engaged owners who use AI themselves; even so, most assume there is a layer of employee use they cannot see.

How do they keep track? Mostly, they don't — at least not with tools. Asked how they check whether employees use AI, 36% say they simply trust employees to disclose it, 30% use software to detect it, 16% say they can tell from the work itself, and 19% do not monitor at all. So for a majority of owners, oversight of employee AI use rests on trust or inference rather than on any system designed to see it. That mirrors the report elsewhere in this series showing that most workplaces lack a clear AI policy — and it means the "shadow AI" phenomenon is not merely that workers use AI quietly, but that the people running these businesses have little practical visibility into how, or how much, their own staff rely on it. Adoption has run ahead of governance on Main Street just as it has in larger organizations.

Comparative Context

The owner's-eye "shadow AI" figure squares with the worker's-eye view from the companion AI@Work shadow-AI analysis, where 70% of employees said their employer was at least somewhat aware of employee AI use and 84% believed the boss knew about their own use. Owners are a shade less confident than workers assume — two-thirds suspect use they can't see — but both sides describe the same reality: AI use that is broadly sensed yet only loosely tracked.

Conclusion

Among the small-business owners who take a survey like this one, AI has clearly arrived: most see it as relevant, nearly all of those already use it, a majority budget for it, and most plan to spend more. That enthusiasm, set against the low single-digit adoption rate federal establishment data report for all small firms, marks these owners as the leading edge of Main Street's AI adoption. But the leading edge is not the same as a settled practice. The spending buys off-the-shelf tools rather than durable systems; the cost savings that usually justify business technology have not clearly materialized; the shortage of AI-skilled staff bites harder the more a business commits to AI; and most owners cannot actually see how their own employees use the technology, relying on trust where larger firms would use policy and tools.

The through-line is a gap between adoption and capacity. Small businesses can buy AI as easily as anyone — a subscription is a subscription — but they are least equipped to staff it, measure it, and govern it. Closing that gap, more than adoption itself, is what will determine whether AI becomes a genuine advantage for small businesses or simply another cost of keeping up. On the current evidence, their owners are betting on the tools while the harder work of staffing and oversight is still ahead of them.

Appendix A — Methods

Data source. AI@Work survey — CHIP50 Wave 38.6 (Civic Health and Institutions Project / COVID States Project). Total unweighted n = 23,462; weighted n = 23,485. Fielded June 21–July 13, 2026, with an oversample of students and adults 18–22; it is nonetheless a broad adult sample. Estimates are weighted to U.S. Census population targets. Data collection was ongoing at the time of this extract — figures reflect responses through July 13, 2026 and are preliminary.

Universe and bases. The report describes respondents who own a small business that employs non-family (screener bus_own, not available in this extract). Every owner received bus_own_ai (is AI relevant), so its base (unweighted n ≈ 1,241) anchors the owner universe and the bus_own_barrier item (n ≈ 1,238, all owners). The AI-detail items were asked only of owners whose business uses/engages AI: bus_own_ai2 (n ≈ 746), ai_op_cost (n ≈ 748), ai_spend (n ≈ 748), ai_invest (n ≈ 714), ai_monitor (n ≈ 747), bus_own_know (n ≈ 747). The spending-composition item ai_spend2 was asked only of owners who allocated an AI budget (n ≈ 440). Because these bases run from a few hundred to ~1,240, the report is descriptive topline only and no person-level demographic breakdowns (party, education, etc.) are reported — cells would be too small. The one breakdown that is reported is by firm size (bus_emp — see below), on the full owner base and collapsed to three groups so cells stay adequate (Table 9).

Comparability caveat. This is a population survey of self-identified owners, not an establishment survey; it is not comparable to firm-level adoption rates such as the Census BTOS 8.8% figure. It describes the attitudes and reported behavior of these owners, weighted to population targets on individual demographics.

Coding. Wording is verbatim in Appendix C, mapped from the authoritative AI@Work survey codebook (these items are not in the machine-readable wording registry). Two ordinal magnitude items were direction-checked in-data against how much the business uses AI (bus_own_ai2): ai_invest (5 = increase significantly … 1 = decrease significantly, 0 = will not spend) — confirmed, its weighted mean rises from 3.04 among light-use businesses to 4.17 among heavy-use ones; and bus_own_barrier (5 = strongly agree … 1 = strongly disagree) — confirmed, mean rises from 2.28 to 3.93 across the same gradient. The remaining owner items (bus_own_ai, bus_own_ai2, ai_op_cost, ai_spend, ai_spend2, ai_monitor, bus_own_know) are labeled category sets reported per the codebook. bus_emp (employee count) is coded 0 = "0–1" … 7 = "1,000+" in the codebook; the data return codes 1–7 ("2–9" through "1,000+") with the "0–1" band absent by the employs-non-family screener. The earlier offset concern is resolved by confirming the data files return raw stored codes for other 0-indexed owner items (bus_own_ai2, ai_invest), so bus_emp values map directly to codebook codes; a firm-size breakdown (Table 9) is reported on that basis, collapsed to three groups (2–9 / 10–49 / 50+) with the direction of the gradients robust to any residual one-band labeling shift.

Estimation. Cross-tabulations are weighted shares. Select-all items (ai_spend2) report the share selecting each option and do not sum to 100%. The by-use gradients for ai_invest and bus_own_barrier are descriptive weighted means, not modeled effects. Cells with unweighted n < 10 are suppressed.

Reproducibility. Figures map to Appendix B tables; values copied verbatim from the underlying tabulations. External benchmark fetched and verified from the SBA Office of Advocacy spotlight (see Appendix C).

Appendix B — Data tables

At least "some" = 88.3%.

Planned increase (significantly + slightly) = 62.0%; decrease = 7.7%.

Agree (strongly + agree) = 42.2%; disagree = 28.7%. Weighted mean by business AI use (bus_own_ai2, 1 = strongly disagree … 5 = strongly agree): uses AI "not at all" 2.28, "not so much" 2.93, "some" 3.38, "a lot" 3.93 — agreement rises with use.

Suspect unsanctioned use (frequently + occasionally) = 67.3%.

Both gradients are monotonic and interpretation-robust: AI relevance and the felt AI-skills shortage each rise steeply with headcount. (Cells for the largest individual bands are thin — 500–999 and 1,000+ have unweighted n ≈ 55–93 — so bands are collapsed to three groups here; the underlying seven-band figures run relevance 42→84% and barrier-agreement 26→73% from smallest to largest.)

Appendix C — Question wording (verbatim)

• bus_own (screener; not available in extract) — "Do you own a small business that employs non-family?" (Yes; No)
• bus_emp (owners) — number of people employed. Codebook bands: 0 = 0–1; 1 = 2–9; 2 = 10–24; 3 = 25–49; 4 = 50–99; 5 = 100–499; 6 = 500–999; 7 = 1,000+. Data return codes 1–7 (band 0 absent by the screener). Collapsed for reporting into 2–9 / 10–49 / 50+.
• bus_own_ai (owners) — "Is AI relevant to your business?" (Yes; No)
• bus_own_ai2 (owners for whom AI is relevant) — "How much does your business use AI?" (A lot; Some; Not so much; Not at all; Not sure)
• ai_spend (AI-using owners) — "Have you allocated budget or funds for AI?" (Yes; No; Not sure)
• ai_spend2 (owners who allocated a budget; select all) — investment composition (Individual subscriptions; Enterprise licenses / team plans; Custom development, hardware, or AI staff; Not sure; None of the above)
• ai_invest (AI-using owners) — "Over the next 12 months, your AI spending will…" (Increase significantly; Increase slightly; Stay the same; Decrease slightly; Decrease significantly; Will not spend)
• ai_op_cost (AI-using owners) — "Has AI changed your operating costs?" (Increased; Decreased; No change; Too early to tell; Not sure)
• bus_own_barrier (owners) — "How much do you agree or disagree: a lack of staff with AI skills is a barrier?" (Strongly agree … Strongly disagree)
• bus_own_know (AI-using owners) — "Do you suspect that your employees use AI without your knowledge?" (Yes, frequently; Yes, occasionally; No, they don't use AI without my knowledge; No, I don't think they use AI at all)
• ai_monitor (AI-using owners) — "Do you check whether employees use AI?" (Software to detect it; Trust employees to disclose it; Can tell from the work; Do not monitor)

External source cited:

• SBA Office of Advocacy, Research Spotlight — AI in Business: Small Firms Closing In (Sept 24, 2025), summarizing Census Bureau Business Trends and Outlook Survey data: ~8.8% of small businesses (<250 employees) using AI as of August 2025; ~50% of small firms using AI made no related investment; relevance is the leading barrier. https://advocacy.sba.gov/wp-content/uploads/2025/09/Research-Spotlight-AI-in-Business-Small-Firms-Closing-In_-092425.pdf