How small businesses
are actually using AI.
Adoption headlines range from 17% to 76% depending on who’s counting. We pulled the U.S. Census Bureau, JPMorgan Chase Institute, the SBA, McKinsey, Salesforce, Goldman Sachs, and 20+ other primary sources to find out what’s real.
Scroll to explore“76% of small businesses report using AI — but only 14% say it’s fully embedded in their core operations. That gap is where the real productivity divide lives.”
— Goldman Sachs 10,000 Small Businesses Survey, 2026Why this matters more than picking “the right” number
Every AI adoption statistic you’ll ever read is a product of its definition, and definitions vary more than most coverage admits. The Census Bureau’s Business Trends and Outlook Survey asks about active use “in any business function” through a large, biweekly, nationally representative sample — it’s built to catch sustained, real operational use and specifically designed to exclude a one-time test of a free tool. Goldman Sachs asked a much broader question of a different population: “have you used AI in your business,” full stop, with no requirement for frequency or depth.
Neither survey is wrong. They’re answering different questions, and treating them as competing claims about the same fact is the single most common error in AI adoption coverage. The useful takeaway for a business owner isn’t “which number is true” — it’s recognizing that your own honest self-assessment probably falls somewhere between these two poles, and that the meaningful milestone isn’t crossing into the “have used AI” category, it’s moving toward the “actively, measurably use it” one.
The findings that matter most
1.2×
The small-vs-large enterprise AI adoption gap has narrowed from 1.8× in Feb 2024 — closing faster than any prior technology cycle.
SBA Office of Advocacy, Sept 2025
91%
of AI-using SMBs report a revenue boost — but only 8% reach advanced, strategic adoption.
Salesforce / Forbes-SMB Group
82%
of the smallest non-adopters (under 5 employees) say AI simply doesn’t apply to them.
SBA Office of Advocacy
Marketing → HR
is the consistent order AI lands inside a small business, regardless of industry — lowest-integration-risk functions first.
HubSpot, Salesforce, Deloitte
14%
say AI is fully embedded in core operations — the real maturity ceiling right now.
Goldman Sachs, 2026
63–68%
projected adoption by mid-2027 if current “plan to adopt” conversion rates hold.
Stacc analysis of SBA/Census data
Small businesses are closing the enterprise gap
Regardless of methodology, every source agrees on direction and speed. Watch the lines converge.
Dark line = large firms (250+ employees). Light line = small firms. Source: SBA Office of Advocacy, September 2025.
Why the gap is closing — and why that’s unusual
Every prior wave of business technology — ERP systems, cloud computing, even basic CRM software — followed the same adoption pattern: large enterprises with dedicated IT budgets and technical staff moved first, and small businesses followed years later, if they adopted the technology at all. AI is breaking that pattern, and the reason is structural rather than cultural. The technology arrived as a consumer-priced subscription rather than an enterprise procurement process. A small business owner can start using a capable AI tool for less than the cost of a single part-time employee’s weekly wages, with no IT department, no server, and no multi-month implementation.
That said, the 1.2× gap that remains is not nothing — large firms still adopt AI more, just not overwhelmingly more. What’s plateaued isn’t large-firm interest, but large-firm adoption rate itself; big companies with the most obvious enterprise use cases (customer service at scale, large-scale data analysis) largely already adopted early, while the growth curve for small firms is still climbing toward that same ceiling. If both curves continue at their current trajectories, small business adoption could match or exceed large-enterprise adoption within the next 18–24 months — an outcome with no real precedent in prior business technology cycles.
Adoption, sliced three ways
Toggle between business size, industry, and department to see where AI has actually landed.
Source: Service Direct 2025 Small Business AI Report; Thryv, 2025.
Includes businesses of all sizes; small-business-only sector splits are thinner in published research. Sources: Census BTOS; Presenc AI; PIIE; AGC.
Source: HubSpot, Salesforce, Deloitte, Intuit QuickBooks, Service Direct, Zapier, SBE Council.
Why departments adopt in this exact order
The department ordering — marketing, then sales and operations, then customer support, then finance, and HR trailing every time — isn’t random, and it isn’t really about which department benefits most from AI. It’s about integration risk. Marketing content lives largely outside of sensitive business systems: a draft blog post or social caption that’s slightly wrong costs almost nothing to fix. Finance and HR work sits inside systems with real compliance exposure and real consequences for errors — a miscategorized expense or a biased resume screen isn’t a minor inconvenience, it’s a liability.
That risk gradient predicts adoption order better than perceived value does. HR’s 19% adoption rate isn’t because AI has nothing to offer HR — resume screening and onboarding documentation are both plausible, useful applications — it’s that the downside of getting it wrong (a discrimination claim, a compliance violation) is severe enough that most small businesses are waiting for the tools, and their own confidence, to mature further before adopting.
Seven industries, up close
67Healthcare adoption in 2026, up from 38% in 2024
Healthcare
The fastest acceleration of any tracked sector — a 29-point jump attributed to expanding FDA clearance of AI-enabled medical devices (178 in 2025, up from 91 in 2024) and EHR platforms integrating AI assistants directly. For small practices, clinical documentation and AI-powered intake are the clearest early wins.
Tools, challenges, and what’s next
Why now: The jump from 38% to 67% didn’t happen because small practices suddenly became more technical — it happened because the tools got embedded into software they already used. When an EHR vendor ships an AI-powered scribe as a feature update rather than a separate purchase, adoption stops requiring a deliberate decision at all. That’s a structural shift worth noting for any small business watching a category mature: the fastest adoption curves tend to come from embedding, not from standalone product launches.
Where practices are actually using it: Ambient clinical documentation (AI listening to a visit and drafting notes) and automated patient intake are the two use cases with the clearest before-and-after story — front desk staff report meaningfully less time on repetitive data entry, and clinicians report less after-hours charting.
The friction that remains: HIPAA compliance and liability exposure are real, not theoretical, concerns in this sector — a wrong AI-generated note in a patient chart carries different stakes than a wrong AI-generated marketing email. Practices moving fastest are the ones treating AI output as a draft that a clinician reviews, not a final record.
Looking ahead: Expect the adoption curve to keep climbing as more EHR platforms bundle AI natively — the practices likely to lag are smaller independent offices still running older, non-integrated systems where AI has to be bolted on rather than switched on.
Legal
Clio’s 2026 survey found 71% of solo practitioners and 75% of small firms use AI personally — remarkably high. But only about a third report an associated revenue increase, and most haven’t adjusted pricing to reflect AI-driven efficiency. 43% of firms have no formal AI policy at all (8am, 2026).
Tools, challenges, and what’s next
The gap that matters: Legal is the clearest example in this entire report of individual adoption running far ahead of organizational adoption. Lawyers are personally experimenting with AI for research, drafting, and summarization at rates that rival tech companies — but the firm itself, as a business, hasn’t caught up. No governance policy, no updated billing model, no formal training. That gap is where risk concentrates: an associate using an ungoverned AI tool on client work is a malpractice exposure the firm may not even know exists yet.
Why revenue isn’t following usage: Most small firms still bill hourly. If AI cuts the time to draft a contract in half, an hourly biller captures less revenue for the same output — the opposite of the incentive a firm needs to invest further. Firms that have found revenue gains are disproportionately the ones that shifted at least some work to flat-fee or value-based pricing, where efficiency gains translate directly into margin instead of into a smaller invoice.
What’s realistic next: Expect governance to catch up over the next 12–18 months, driven less by internal initiative and more by malpractice insurers and state bar associations starting to ask firms directly what their AI policy is.
75%% of small law firms report personal AI use
31%Construction — the lowest adoption of any major sector
Construction
Explained less by resistance and more by structural mismatch — field-based, project-driven work with limited digitized data doesn’t fit most current AI tools, which were built for office-based, text-heavy workflows. Source: Associated General Contractors, 2025.
Tools, challenges, and what’s next
Why the mismatch is real, not just cultural: Most AI tools on the market today are built around text and structured data — documents, spreadsheets, tickets. Construction work generates a different kind of data: photos of a job site, verbal instructions on a crew call, handwritten change orders. Until recently, there wasn’t much AI could do with that. The tools that are gaining traction — computer vision for site progress tracking, AI-assisted takeoff and estimating software — are the ones built specifically around construction’s actual data shape, not general-purpose tools awkwardly applied to it.
Where the ROI case is strongest: Bid estimating and scheduling are the two areas small contractors report the clearest time savings, since both are bottlenecked by a single skilled person doing manual work that AI-assisted tools can accelerate without needing to touch anything happening physically on-site.
Realistic outlook: Construction is unlikely to close the gap with tech or financial services anytime soon — the physical, distributed nature of the work is a genuine structural barrier, not a temporary lag. The more useful frame for a contractor isn’t “when will construction catch up” but “which specific back-office task — estimating, scheduling, paperwork — is worth automating first.”
Retail
Adoption concentrates where the revenue connection is direct — recommendations, dynamic pricing, forecasting. AI-powered dynamic pricing delivers 5–10% margin gains with a 6–12 month payback, yet fewer than 15% of retailers use it (McKinsey / Alhena AI) — real unrealized upside for first movers.
Tools, challenges, and what’s next
Why retail moves faster when it moves: Unlike a lot of the sectors in this report, retail AI use cases have a direct, easily measured line to revenue — a better product recommendation is a bigger cart, a smarter reorder point is less dead stock. That directness is why retail adoption, while not the highest overall, tends to stick once it starts: the business case doesn’t require faith, it requires a dashboard.
The under-adopted opportunity: Dynamic pricing is the clearest example of a proven use case still sitting mostly unused among small retailers — the 5–10% margin gain McKinsey and Alhena AI both point to isn’t a projection, it’s already measured among the minority of retailers running it. The barrier isn’t proof of value; it’s that most small retail operators have never set up pricing rules beyond a simple markup, so the leap to dynamic, demand-responsive pricing feels bigger than it is.
What’s next: Expect inventory forecasting to be the next use case to cross from “early adopter” to “standard practice” for small retailers, as more point-of-sale and e-commerce platforms bundle basic AI forecasting directly into tools retailers already use daily.
5–10%% margin improvement from AI dynamic pricing
52%Manufacturing adoption, grounded in measurable outcomes
Manufacturing
Predictive maintenance monitoring vibration and temperature, computer vision catching production-line defects, and demand-forecasting improving inventory planning — tangible, durable use cases rather than experimental ones.
Tools, challenges, and what’s next
Why manufacturing adoption is durable, not trendy: Every major manufacturing AI use case ties to a number a plant manager already tracks — downtime hours, defect rate, inventory turns. That’s a meaningfully different adoption pattern than sectors where AI use is more exploratory: manufacturers don’t tend to adopt and then abandon, because the tools are answering questions the business was already measuring before AI existed.
Where small manufacturers hit friction: Predictive maintenance requires sensor data the equipment has to actually produce — older machinery without IoT sensors can’t feed a predictive model no matter how good the software is. That’s the real barrier for smaller shops: it’s often a hardware and instrumentation gap before it’s a software gap.
What’s next: As sensor retrofitting gets cheaper, expect predictive maintenance to spread beyond the larger manufacturers who could afford full IoT builds, into smaller shops that can now add targeted sensors to just their highest-risk equipment rather than instrumenting an entire facility.
Hospitality
Roughly one-fourth of travel and hospitality companies have adopted chatbot technology specifically. The gap is real first-mover opportunity — administrative automation and scheduling need minimal setup and directly address staffing pressure.
Tools, challenges, and what’s next
Why this sector lags despite an obvious use case: Guest-facing chatbots are one of the most intuitive AI applications in any industry — answering “what time is checkout” doesn’t require deep technical sophistication. The lag isn’t about proving value; hospitality businesses tend to run on thin margins and thinner administrative staff, leaving little bandwidth to evaluate and implement even a straightforward tool.
Where the real pressure point is: Staffing shortages, not guest experience, are the actual driver behind hospitality’s growing AI interest — AI handling routine guest questions and reservation changes frees limited front-desk staff for the interactions that genuinely need a human.
What’s next: Expect adoption to track closely with major property management and booking platforms bundling AI features natively, the same embedding pattern seen in healthcare — small hospitality operators are far more likely to adopt AI that arrives inside software they already use than to go shopping for a standalone tool.
~25%% chatbot adoption in travel & hospitality
33%Small education & training businesses using AI
Education
Concentrated in content creation, grading assistance, and student engagement tools (National Center for Education Statistics, 2025) — also one of the fastest-accelerating sectors overall.
Tools, challenges, and what’s next
Why this sector is accelerating: Small education and training businesses — tutoring services, corporate training providers, course creators — sit closer to content production than most small businesses, and content production is exactly where generative AI shows its clearest advantage. Drafting lesson plans, generating practice questions, and personalizing feedback are all tasks that map directly onto what large language models do well.
The tension worth naming: Grading assistance in particular raises a legitimate quality question — AI-assisted feedback needs a human check before it reaches a student, the same “draft not final record” principle seen in healthcare documentation. Providers moving fastest treat AI as accelerating a human educator’s work, not replacing their judgment.
What’s next: Expect personalized, adaptive content — tools that adjust difficulty and pacing to an individual learner — to be the next wave, building on the content-creation and engagement tools already in wide use today.
The AI stack, orbiting one hub
Hover each node — sizes reflect relative adoption share among AI-using small businesses.
Categories aren’t mutually exclusive — businesses commonly appear in more than one. Sources: HubSpot, Salesforce, Deloitte, Intuit QuickBooks, 2025.
The return is real — for the businesses that stick with it
Salesforce found 91% of SMBs using AI report a revenue boost, and 86% saw improved margins. Training matters more than the tool: 4–8 hours of structured training produces 2.3× higher task completion rates (Deloitte).
Sources: Salesforce SMB Trends (2024); Intuit QuickBooks (2025); Deloitte (2025); SBE Council (2026).
Why training outperforms the tool itself
The single most consequential finding in the entire ROI dataset isn’t a percentage — it’s the comparison between trained and untrained deployment. Deloitte’s 2.3× productivity multiplier for businesses that invest 4–8 hours of structured training isn’t a marginal improvement; it’s the difference between AI being a novelty a few employees dabble with and AI actually changing how work gets done. Most of the ROI variance across small businesses traces back to this single variable more than to which specific tool was purchased.
What “training” actually means in practice for a small business rarely resembles a corporate learning program — it’s closer to a structured internal walkthrough: here’s the tool, here’s the three tasks we’re using it for, here’s what a good output looks like versus a bad one, here’s who to ask when it gets something wrong. Businesses that skip this step tend to see AI used inconsistently by one enthusiastic employee while the rest of the team ignores it, which shows up in the data as underwhelming aggregate ROI even when the tool itself works well.
The business implication: if you’re budgeting for a new AI tool, budget time alongside it. A cheaper tool with real onboarding will very likely outperform a more expensive tool dropped into the business with no structured rollout.
It isn’t cost. It’s this.
Click to expand each barrier — ranked by how often it’s cited among non-adopters.
82%“Not applicable to my business”
Among non-adopters with fewer than five employees, this is by far the dominant reason cited. It drops sharply as business size increases — a strong signal this is an education gap, not a genuine mismatch. (SBA Office of Advocacy)
~46%Lack of technical expertise
The second-tier barrier cluster is about confidence, not access. (Goldman Sachs 10,000 Small Businesses)
29%Tool-choice overwhelm
More than 8,000 AI products are now marketed to small businesses, making evaluation genuinely overwhelming. (Salesforce)
22%Data privacy concerns
Highest in healthcare, legal, and financial services, where regulatory exposure creates additional risk. (Deloitte)
18%No time to evaluate tools
The dominant barrier for micro-businesses, where the owner performs most operational tasks. (JPMorgan Chase Institute)
5%See no value in AI at all
Genuine philosophical opposition is rare — most non-adopters are blocked by practical barriers, not skepticism. (U.S. Census Bureau)
The pattern underneath all six barriers
Line up all six barriers and a clear pattern emerges: they get less severe, not more, as you move down the list. “Not applicable” (82%) reflects a genuine misunderstanding of what AI can do. “Lack of expertise” (46%) and “tool-choice overwhelm” (29%) are about confidence and time, not capability. “No time to evaluate” (18%) is a resource constraint that shrinks as tools get easier to trial. And “see no value” (5%) — the barrier that would actually justify not adopting — is the rarest response of all.
That ordering has a direct implication for how AI vendors, and AI-focused publications like this one, should be talking to small business owners: the primary obstacle isn’t proving that AI works. It’s translation — showing a specific owner, in their specific industry, exactly which task in their specific business AI would touch. Generic “AI will transform your business” messaging speaks past the actual barrier. Specific messaging — “here’s what AI does for a landscaping company’s scheduling” — speaks directly to it.
From chat to autonomous action
Generative AI use jumped from 37% to 72% of organizations between 2023 and 2025 (McKinsey). AI agents — systems that plan and execute multi-step tasks — are the newer, less mature frontier.
Source: McKinsey State of AI, 2025. Gartner projects more than 40% of agentic AI projects will be canceled by 2027, largely due to governance gaps rather than the technology failing — start narrow, keep a human in the loop.
Why “40% will be canceled” isn’t actually a warning against agents
Gartner’s cancellation projection gets cited constantly as a reason to be cautious about AI agents, and it should be — but the detail that usually gets dropped is why those projects fail. Gartner’s own analysis attributes the majority of cancellations to inadequate risk controls, unclear success metrics, and rising costs — governance and planning failures, not the underlying technology failing to perform the task it was given. That distinction matters enormously for a small business owner deciding whether to touch agents at all.
The practical translation: an agent that drafts and sends marketing emails without a human checking them first is a governance failure waiting to happen, regardless of how capable the underlying model is. An agent that drafts marketing emails for a human to review before sending captures most of the time savings with almost none of the downside risk. The businesses in the 23% “scaling in production” bracket overwhelmingly built that human checkpoint in from day one rather than adding it after something went wrong.
What this means for a first agent project: pick a task where a wrong output is annoying, not damaging — a draft, not a decision — and expand the agent’s autonomy only after it’s proven reliable on that narrow, low-stakes task.
Where most small businesses actually stand
Hover each tier. The overwhelming majority sit in the bottom two.
Experimental
Testing a free tool occasionally, no defined workflow.
~25%
Exploring
Regularly using 1–2 tools for a specific task without full commitment.
~51%
Operational
AI embedded in recurring workflows with measured savings.
~23%
Strategic
Integrated across multiple functions, intentional strategy.
~8%
Sources: Forbes / SMB Group; Thryv’s “AI explorer” self-identification; MIT (76% of AI-using orgs limited to 1–3 use cases).
Why so few businesses reach the top tier — and what actually moves you up
The jump from “Exploring” to “Operational” is the hardest transition in this model, and it’s worth understanding why: it’s the point where a business has to stop treating AI as an occasional convenience and start treating it as infrastructure the business would notice if it disappeared. That requires something most small businesses never do for any tool — measuring a before-and-after baseline, documenting how the workflow actually works, and making sure more than one person knows how to run it.
MIT’s finding that 76% of AI-using organizations remain limited to just one to three use cases is the clearest evidence of where businesses get stuck: they successfully adopt AI for one task, see it work, and then never systematically expand from there. The businesses that reach “Strategic” maturity are almost never the ones that adopted the most tools — they’re the ones that treated their first successful use case as a template to repeat deliberately in a second and third function, following something close to the roadmap in the next section rather than adopting opportunistically.
A realistic 6-month roadmap
Swipe through — based on the pattern among businesses that reached operational or strategic maturity.
Month 1
Pick one painful task
Not “adopt AI generally” — one specific bottleneck, like drafting marketing content or triaging inquiries.
Month 2
Trial one tool, measure a baseline
Track current time or cost before introducing a tool, so improvement is measurable.
Months 3–4
Train the team, not just the tool
4–8 hours per employee is linked to 2.3× higher task completion (Deloitte).
Month 5
Add a second tool
Only after the first shows measurable ROI — how the median 5-tool stack gets built without waste.
Month 6
Reassess and set a real budget
Use our ROI Calculator to formalize the case for continued investment.
← Swipe or scroll →
Outlook Through 2027
0%
Projected small business AI adoption by mid-2027 — if current “plan to adopt” conversion rates (historically 55–65% for small business tech) hold against today’s 31% of non-adopters with active plans.
The realistic caveat: activity isn’t the same as impact. McKinsey found only 39% of organizations report any enterprise-level financial impact from AI, and PwC found 56% of CEOs report zero measurable ROI — almost entirely among businesses that layered AI onto existing processes rather than redesigning around it. The winners through 2027 won’t be the ones who adopted first. They’ll be the ones who matched a specific tool to a specific problem, measured the result, and built from there.
Three predictions worth planning around
Not certainties — reasoned extrapolations from the trends documented throughout this report.
1. The enterprise-vs-SMB gap likely closes entirely, or inverts. If small business adoption continues climbing at its current rate while large-firm adoption holds near its current plateau, small businesses could reach parity with — or exceed — large enterprise adoption rates within 18–24 months, an outcome with no precedent in prior business technology cycles.
2. Agent adoption follows generative AI’s curve, roughly two years behind. Generative AI took from 2023 to 2025 to move from 37% to 72% organizational use. Agents are earlier on a similar-shaped curve today; if the pattern holds, expect agent “scaling in production” to move from today’s 23% into the 50–60% range by 2027–2028, assuming the governance failures Gartner warns about get addressed rather than ignored.
3. The “not applicable” barrier collapses faster than any other. Because 82% of the smallest non-adopters cite genuine misunderstanding rather than a real mismatch, this is the barrier most susceptible to simple exposure — as more small business owners personally see a peer’s specific, concrete use case, expect this figure to fall faster than skill- or cost-related barriers, which require more than just awareness to resolve.
Every number, traceable
No statistic on this page was invented. Click any source for what it actually measures.
This report synthesizes publicly available data from government agencies, academic research institutions, and industry surveys published between late 2024 and mid-2026. Where sources use different definitions of “AI adoption” — the recurring theme throughout this report — that difference is stated explicitly rather than smoothed into one tidy number. Adoption data moves quickly; figures reflect the most recent published wave of each source as of this report’s publish date and should be expected to shift as newer survey rounds are released. We treat government and transaction-based sources (Census, JPMorgan Chase Institute) as the most methodologically conservative baseline, and vendor surveys as directionally useful but subject to the self-selection biases inherent in any opt-in survey of a company’s own customer base.
U.S. Census Bureau (BTOS)
JPMorgan Chase Institute
SBA Office of Advocacy
McKinsey & Company
Salesforce
Goldman Sachs
Clio & 8am
Thryv, SBE Council, Intuit, Deloitte
See also our How We Test AI Software and Review Methodology pages for how this research connects to our hands-on AI software reviews.
Quick answers
Practical questions people have about small business AI adoption.
01What percentage of small businesses actually use AI?
It depends on the definition. Census Bureau BTOS puts strict production use at roughly 17–20%. Thryv found 55% self-reporting any use. Goldman Sachs found 76%. All three are accurate; they measure different things.
02Which department adopts AI first?
Marketing, consistently, across nearly every survey — the lowest barrier to entry and the fastest visible results.
03Which industries have the lowest adoption?
Construction, at roughly 31% (Associated General Contractors), followed by hospitality — field-based, project-driven work adopts more slowly than office-based work.
04Is AI adoption actually delivering ROI?
For businesses past one-time experimentation, yes — 91% report a revenue boost (Salesforce). Gains concentrate among businesses that matched a specific tool to a specific workflow.
05What’s the single biggest barrier?
Perceived irrelevance — 82% of non-adopters under 5 employees say AI simply doesn’t apply to them (SBA Office of Advocacy).
06How many AI tools does the average small business use?
A median of five — typically a general assistant plus tools for customer service, marketing, scheduling, and analytics.
Want research like this in your inbox?
One practical AI idea for your business every Friday, tested by us first. No hype, no spam.