How are small companies using AI?

Adoption Has Tipped: The State of Small Business AI

Two years ago, AI in a small business was an experiment. Today it's close to the default: roughly 60% of U.S. small businesses now use AI in their operations, more than double the 2023 rate, according to U.S. Chamber of Commerce data. A separate Goldman Sachs survey puts the figure even higher, with 68% of small business owners reporting they already use AI, up from 51% two years earlier.

The newest businesses show the sharpest shift. JPMorgan Chase Institute cohort data found that paying for an AI tool in a company's first month was stable at roughly 1.2–1.7% for businesses founded between 2019 and 2023 — then jumped to 4.1% for 2024 cohorts and 6.5% for 2025. That's about five times the 2019 rate, a sign that AI is becoming part of the startup kit rather than a later upgrade. The adoption gap between small and large businesses also narrowed from 1.8x to 1.2x in about 18 months, faster than the broadband adoption curve.

But the headline numbers hide a contradiction most owners feel firsthand. Everyone says AI is everywhere, yet only 25% of small businesses actually use it in daily operations. Per the NSBA survey, 76% are using or exploring AI, but the single biggest group — 51% — are what researchers call "AI Explorers": interested, experimenting, and not yet committed.

Why the stall? The barriers are practical, not philosophical:

• 48% struggle to choose the right tools from a crowded market
• 46% worry about data privacy and security
• 41% cite a lack of technical expertise

As the NSBA data puts it, small businesses "want to see clear value before they commit." The remaining gap, sources agree, is about training and demonstrated ROI — not resistance to the technology.

That gap is exactly what SkillzBench is being built to address: a free, independent directory of AI agent skills where every skill is security-scanned and benchmarked with a published Lift score, so owners can see proven value before wiring anything into their workflow. If you want to follow along as it launches, the waitlist is open now.

The takeaway for 2026 planning: adoption has tipped, but implementation hasn't. The winners won't be the businesses that tried the most AI tools — they'll be the ones that picked one workflow, proved the lift, and made it stick.

The Explorer Gap: Why Interest Isn't Turning Into Daily Use

Small business AI adoption has a strange shape: interest is everywhere, but daily use is not. According to NSBA survey data, 76% of small businesses are using or exploring AI, yet only 25% use it in day-to-day operations. The 51% in between — the "AI Explorers" — haven't seen enough demonstrated value to commit.

That gap isn't resistance. It's a stack of practical blockers. Goldman Sachs survey data identifies the three most-cited barriers:

• Tool selection — 48% struggle to choose the right tools.
• Data privacy and security — 46% worry about how their data is handled.
• Lack of technical expertise — 41% don't feel equipped to apply AI to their business.

The smallest firms are the furthest behind. Just 5.8% of businesses with 1–4 employees use AI, and SBA BTOS data shows 82% of very small firms believe AI simply doesn't apply to them. They're not wrong to be skeptical — they've seen hype cycles before, and no one has shown them a measurable result inside a business like theirs.

What Explorers actually want is proof, not persuasion. The same research finds 74% want clearer ROI evidence before committing and 73% want easier-to-use tools, with practical training ranked as the top support need. That's precisely the gap SkillzBench is being built to close: an independent, free directory that verifies agent skills, benchmarks each one in a sandbox with and without it, and publishes the Lift in percentage points — so a five-person firm can see exactly what a skill does before trusting it with a workflow.

What's Actually Working: Marketing First, Then Everything Else

The businesses seeing real returns from AI aren't experimenting broadly — they're picking one workflow, measuring it, and letting the results speak. According to small business AI research, "the key to ROI is starting with a specific, measurable workflow rather than broad experimentation." That's exactly how the successful adopters are doing it.

Marketing is the proven entry point. Content generation is the most popular AI use case among small businesses, and the time savings are immediate: tasks that once took four hours now take under an hour. HubSpot data suggests small teams save 5–15 hours per week on marketing tasks alone — which is why marketing and customer engagement is the top priority for 77% of small businesses, per Salesforce figures.

Customer service is the natural second step. Chatbots now handle 40–60% of routine inquiries without pulling staff away from higher-value work, according to aggregated industry statistics. For a lean team, that's the equivalent of adding a coverage shift without adding headcount.

Beyond marketing and support, concrete workflow examples are emerging across service businesses:

• Chatbot "clones" for consultants and coaches: service-based entrepreneurs upload course materials, blog posts, and FAQs to train chatbots that handle client interactions, scheduling, and preliminary consultations around the clock, per U.S. Chamber guidance.
• AI-generated SOPs from screen recordings: training tools auto-generate step-by-step procedures from a recorded screen session, cutting training costs and speeding up onboarding.
• Competitive intelligence: AI tools analyze competitors' SEO strategies, social engagement, and product positioning faster than manual research ever could.
• Cash flow forecasting: owners increasingly want tools that spot customer trends and guide resource decisions, according to NSBA survey findings.

The pattern across all of these is the same: pick a workflow with a clear baseline, apply AI, and measure the difference. Owners who follow this path report the payoff — roughly 80% of small businesses using AI say it enhances their workforce rather than replacing it, and 74% of AI users plan to grow in 2025, according to Goldman Sachs survey data.

For those ready to go further, agent skills — instruction packages that extend what AI agents can do — can automate the rest of a chosen workflow. SkillzBench is building a free directory where every skill is security-scanned and benchmarked for measurable lift, so you can see what actually works before you commit. Join the waitlist to be first in line when it launches.

Your First Workflow: A Pick-Measure-Apply-Measure Playbook

Don't start with a big AI strategy. Start with one workflow, one timer, and one number.

The evidence says narrow beats broad: research on small business AI puts it plainly — "the key to ROI is starting with a specific, measurable workflow rather than broad experimentation." That matters because 51% of small businesses are "AI Explorers" who haven't committed, and 74% of them say they want clearer ROI evidence before they do.

Here's the four-step playbook.

Pick one workflow — marketing is the proven entry point. Content generation is the most popular small business use case, and marketing is where ROI shows up fastest: tasks that once took four hours now take under one hour. Other candidates from U.S. Chamber reporting include auto-generating SOPs from screen recordings or running competitive-intelligence scans.

Measure your baseline. Before touching any tool, time the workflow as you do it today. If it's a weekly blog post, a customer email sequence, or an FAQ chatbot, log the hours. Without this number, you can't prove anything later.

Apply one AI tool or agent skill. This is where most owners stall — 48% struggle to choose the right tools and 46% worry about security. Agent skills (instruction packages that give coding agents and chat-app agents new capabilities) can automate individual steps of your workflow, but quality varies wildly: published research found roughly 26% of skills carry at least one vulnerability. This is the problem SkillzBench is being built to solve — a free directory that security-scans every skill, pins versions, and benchmarks each one in a sandbox with and without the skill, publishing the Lift in percentage points so you can see demonstrated value, not vendor promises.

Measure the lift, then expand. Re-time the same workflow. If you save three hours a week on one task — consistent with HubSpot's finding that AI users save 5–15 weekly hours on marketing alone — you have your proof. Expand to the next workflow only after the first one shows numbers.

The pattern matters more than any single tool. Owners who commit to AI aren't the ones who experimented everywhere; they're the ones who measured something specific and saw it work. Once one workflow proves out, the same pick-measure-apply-measure loop applies to customer service, training, and forecasting.

If you want a head start on step three, SkillzBench's free skills directory and MCP connector are designed to let you browse benchmarked skills, toggle exactly what your agent should see, and test with real Lift scores — everything free, with a waitlist now open.

• Pick one measurable workflow (marketing recommended)
• Time the baseline before adding AI
• Apply one vetted tool or agent skill
• Measure the lift in hours saved
• Expand only after the numbers prove out

Choosing Tools You Can Trust Without a Technical Team

If you don't have an engineer on staff, every AI tool decision is a leap of faith — and the data shows that leap is where most small businesses stall. Nearly half struggle to choose the right tools, and 46% worry about data privacy and security, according to a Goldman Sachs survey of small business owners. Trust, not technology, is the real gatekeeper.

The stakes are higher than most owners realize. Agent skills — the instruction packages that extend tools like Claude Code, Cursor, and Gemini CLI — are proliferating across GitHub, but nobody is checking them before you install them. Published research found that roughly 26% of agent skills carry at least one vulnerability. For a non-technical team, that's like hiring a contractor without checking references.

The good news: you don't need technical expertise to ask the right questions. Before adopting any AI tool or skill, check:

• Security — has the tool been scanned for vulnerabilities, and is the version you're getting pinned and traceable?
• Licensing — are you legally allowed to use it in a commercial context, or does the license restrict business use?
• Proven results — has the tool been tested against a baseline, with real numbers showing what it actually improves?
• Maintenance — is it actively updated, or will it break silently when the underlying platform changes?

This is exactly the verification gap SkillzBench is being built to close. The free directory will pin every skill version to a commit SHA, run license checks and security scans, and assign a 0–100 trust score based on security, quality, license, maintenance, and community signals. Crucially, benchmark results — the Lift a skill delivers, measured in percentage points in a sandbox with the same model — are always shown separately from ratings, and money never touches scores.

For small teams, the workflow is simple: browse verified skills, toggle the ones you want into a per-agent loadout, and connect them through a free MCP tool — your agent only sees what you chose. Everything, including the directory, connector, benchmarks, and change alerts, is free.

That matters because the smallest businesses are the most cautious: only 5.8% of firms with 1–4 employees use AI, and industry statistics show 82% of very small firms believe AI doesn't apply to them. Free, verified tools with demonstrated value are precisely what changes that math.

SkillzBench is pre-launch and actively being built. If you want to equip your agents with skills you can actually trust — and see the verification infrastructure the moment it goes live — join the waitlist at skillzbench.com.