AI Readiness and Results

Stuck Testing AI? What Separates Small Firms That Use It

Most small businesses now use AI, but few have built it into how they work. A 2026 survey of 402 US small business leaders shows what separates the two groups: leaders who agree what AI is for, a rule staff can read and a named owner. Here is how to check yours this week.

A signpost with three paper arms pointing different ways beside three large arrows all pointing the same way, standing for a business agreeing what AI is for.

In brief

Small businesses that use AI daily and those stuck testing it mostly have the same tools. In a July 2026 Pax8 survey of 402 US small business leaders, 91 percent of active users said leadership was aligned on AI's role, against 68 percent of experimenters, and users were about twice as likely to have a rule. Answer three questions the same way across the business (what AI is for, what staff may do, who decides) before buying anything else.

Small businesses that use AI every day and small businesses still testing it mostly buy the same tools. What separates them, in a July 2026 survey of 402 US small business leaders, is agreement: leaders who give the same answer about what AI is for, a rule staff can read, and one person who decides. None of the three has to be bought. You can start this week.

What Does the Research Say Separates Users From Testers?

The clearest difference is whether the people at the top agree about what AI is for. Pax8, a company that sells cloud software through IT service providers, runs a quarterly survey of small business leaders. Its Q2 2026 SMB AI Pulse Report, released July 13, 2026, asked 402 US leaders at companies with 5 to 499 employees where they stood. Sixty-one percent said they were actively using AI. Another 29 percent were still experimenting, a group the release describes as "stuck in experimentation."

Then it compared the two groups.

Among businesses actively using AI, 91 percent said leadership was fully or mostly aligned on AI's role. Among experimenters the figure was 68 percent, and among non-users 32 percent. The release calls this "a clear staircase." The rules follow the same steps. About two-thirds of AI users had either a documented policy or informal guidelines. Among experimenters, 34 percent had reached that point, and only 11 percent had a documented policy.

Two cautions before you lean on those numbers. The survey shows that agreement and everyday use travel together; it cannot show which came first. And the experimenters are about 29 percent of 402 people, so their percentages are a sketch, with a wider margin than the 4.9 points quoted for the whole sample.

Other 2026 surveys describe the same crowded middle. Goldman Sachs asked 1,256 US owners in its 10,000 Small Businesses program between January 27 and February 4, 2026. Seventy-six percent were using AI, and yet "Only 14% say AI is fully embedded in their core operations." In the UK, the government's AI Adoption Survey 2026, published October 29, 2026, interviewed 3,503 businesses between January 7 and April 21. A third of them (32 percent) were using at least one AI technology, double the 2025 share, but only 14 percent considered AI tools of high or critical importance to their operations. Its summary is blunt: "Most businesses were still in an early or exploratory stage, with gradual progression towards more advanced use cases."

Plenty of AI, then. Not much that the business would miss if it vanished on Monday.

Why Do Owners Blame Expertise and Cost When the Gap Is Agreement?

Because a missing agreement rarely looks like one from the inside. Pax8 asked the experimenters what was holding them back, and three answers led: lack of internal expertise (28 percent), cost or unclear return on investment (24 percent), and security or privacy concerns (21 percent).

Each of those is real. Each is also what an unanswered question feels like to the person living with it. The table below is my reading of the three barriers, not a finding of the survey.

What testers say is in the way What I usually find underneath The question nobody has answered
"We don't have the expertise" Nobody has been asked to own AI, so nobody has had a reason to become the expert Who decides when it goes wrong?
"We can't see the return" Nobody agreed what AI is for, so there is nothing to measure a return against What is AI for here?
"We're worried about security and privacy" There is no written rule, so every use feels like a risk and the careful people hold back What may I do with it?

Look at the right-hand column. I did not write those three questions for this article. I have been asking them for years.

Which Three Questions Move a Business From Testing to Using?

The three questions of the Readiness Test, answered the same way by everyone who works for you. The Readiness Test is three questions an organization must be able to answer consistently, from the top to the front line, before it can call itself AI ready: What is AI for here? What may I do with it? Who decides when it goes wrong?

Each one lines up with something the Pax8 survey measured, or came close to measuring.

What is AI for here? This is the leadership alignment figure: 91 percent among users, 68 percent among testers. A good answer is one sentence that names a job. "Quotes go out the same day" is an answer. "Efficiency" is a mood. If you have not chosen the job yet, a one-page AI plan is the place to do it.

What may I do with it? This is the policy figure: two-thirds of users with a rule of some kind, a third of testers. The rule can be short. One page a new starter reads in two minutes will do: which tools, what never goes into them, what a person checks before anything leaves the building. I have set out what a small business AI policy needs, and there is a UK template you can copy. If some of your customers are in the EU, check whether the EU AI Act applies to your business before you finish the page.

Who decides when it goes wrong? The survey did not ask this directly. The nearest signal is the expertise barrier, because expertise grows around whoever is given the job. The answer is a name, never a committee, and in most small firms it is someone you already employ with a few protected hours a week.

"Consistently" is the word that does the work in that definition. An owner who can answer all three in the car on the way home has not passed. The business passes when the newest person on the payroll gives the same answers without checking.

The Question to Ask Your Team

Ask three people the first question separately, and write down exactly what each one says. Pick yourself, one manager and whoever joined most recently. Do not warn them, and do not correct them.

Across the business teams I train, the pattern is familiar. The owner thinks the answer is obvious. The operations manager says AI is for cutting admin. The person who runs sales says it is for winning more work. The newest hire says nobody has mentioned it, so she uses it for emails and keeps quiet.

Three different answers do not mean your staff have failed. They mean the business is running three experiments and calling them one. Nobody can finish an experiment they do not know they are part of, which is how a firm stays in testing for a year without anyone deciding to.

My advisory work has included government bodies such as the UK Department for Education, KHDA in Dubai and the Ministry of Education in Kazakhstan, and my training work includes teams of ten. The scale could hardly differ more. In my experience the first question does not change with it.

Do You Need an Outside Partner to Get Unstuck?

Sometimes, but rarely first. Pax8's own conclusion is that small businesses need a technology partner to deploy and govern AI for them. "Most SMBs can't build that alone," says its chief executive, Scott Chasin. Pax8 sells through exactly those partners, so read the advice with that in mind. Its vice president of AI adoption, Chance Weaver, makes the more useful point: "The businesses getting the most from AI are the ones with the strongest support, not necessarily those with the biggest budgets."

Here are the three ways out that owners usually weigh, checked against the sources on November 14, 2026.

Way out What it fixes What it leaves open What the evidence says
Bring in an outside technology partner Setup, security settings, keeping the tools running What AI is for in your business and who decides. A partner cannot answer those for you Pax8 recommends it. Pax8 also sells through such partners
Add more tools or more training Know-how. In the UK survey, limited AI skills, expertise or knowledge affected 67 percent of businesses already using AI The job AI is meant to do and the rule for doing it Goldman Sachs: 73 percent of US owners surveyed said they would benefit from more training and implementation resources
Answer the three questions yourselves first Agreement, a readable rule, a named owner It installs nothing and teaches no skills Pax8: aligned leadership at 91 percent of users against 68 percent of testers. A correlation, not proof of cause

My verdict: answer first, then buy. The first two options work far better once the third is done, because a partner or a trainer who arrives to find three different ideas of what AI is for will build for whichever idea they hear first.

There is an honest exception. If your answer to "What may I do with it?" depends on settings you cannot configure, or on connecting systems that hold customer records, get help before anyone goes further. That is a technical gap, and agreement will not close it. When you do go looking, choosing between a consultant, a course and a workshop is a decision of its own.

If you are bringing your leadership team together to settle the first question, my business keynotes can be paired with a working session for that team.

How Do You Know You Have Stopped Testing?

You have stopped testing when one named job is done with AI by default and somebody would notice if it stopped. Licenses and logins do not count. Nor does a staff survey in which everyone says they use it "sometimes."

Take a hypothetical 20-person lettings agency. For eight months, four people have used an AI assistant to tidy emails, each in their own way. That is testing. Then the owner decides AI is for one job, getting inventory reports to landlords within a day of a check-out. She writes half a page on what may go into the tool, and names the office manager as the person who decides when a report looks wrong. A month later the reports go out next day, every time, and a landlord would ring if they did not. That is use.

Nothing about the technology changed between those two states. Three answers did. If you want to check a job more closely, the Freewheel Diagnostic asks four questions of any job where AI is supposed to help, and I have written separately about measuring whether AI is working and whether your team is really using it.

What Should You Do This Week?

Five short jobs, none of which needs a budget.

  1. Monday. Ask three people the three questions, separately: you, one manager, your newest hire. Ten minutes each. Write down their words, not your summary.
  2. Tuesday. Lay the answers side by side. Wherever they differ, you have found the thing keeping you in testing.
  3. Wednesday. Write one sentence that answers "What is AI for here?" and names a single job. Say it out loud to your managers. If it takes more than a breath, shorten it.
  4. Thursday. Put the rule on one page: which tools, what never goes in, what a person checks.
  5. Friday. Put one name against "Who decides when it goes wrong?" and tell everyone who it is.

Thirty days later, ask the same three people again. If the answers match, you are no longer testing, and you will know far better what, if anything, you need to buy. And if you started late, that matters less than you fear: it is not too late to start, provided the starting is deliberate.

Find Out Which Answer Is Missing

If you would rather see the gaps scored than guess at them, take my free Business AI Readiness Scorecard. It asks twelve questions across six areas, beginning with leadership and direction, takes about five minutes, and gives you a 90-day plan with three priorities. For the fuller picture behind it, read how ready your business is for AI. And if you want to talk it through, get in touch.

Sources checked November 14, 2026.

Sources and further reading

Dan Fitzpatrick is The AI Educator: a Forbes contributor, international keynote speaker and bestselling author who has trained more than 150,000 people across 30+ countries. More about Dan.

Key takeaways

  • In Pax8's Q2 2026 survey of 402 US small business leaders, 61 percent were actively using AI and another 29 percent were still experimenting.
  • Leadership alignment on AI's role was 91 percent among active users, 68 percent among experimenters and 32 percent among non-users. That is a correlation, not proof of cause.
  • About two-thirds of AI users had a documented policy or informal guidelines; among experimenters 34 percent did, and only 11 percent had a documented policy.
  • The barriers testers name (expertise, unclear return, security) often turn out to be three unanswered questions: who decides, what AI is for, and what staff may do.
  • The Readiness Test asks those three questions and counts a pass only when everyone, from owner to newest hire, gives the same answers.
  • Answer the three questions before paying for a partner, more tools or more training, unless the gap is a technical one such as security settings or connected customer records.
  • A business has stopped testing when one named job is done with AI by default and somebody would notice if it stopped.

Frequently Asked Questions

Why is my small business stuck testing AI?

Usually because the people in it do not agree what AI is for. In a July 2026 Pax8 survey of 402 US small business leaders, 68 percent of experimenters said leadership was aligned on AI's role, against 91 percent of active users. Testers were also far less likely to have any rule for staff.

What is the difference between testing AI and using it?

A business is using AI when one named job is done with it by default and somebody would notice if it stopped. Testing is everything short of that: a few people trying tools in their own way, with no agreed purpose, no written rule and nobody who decides when a result looks wrong.

How many small businesses are still experimenting with AI?

About three in ten, on one measure. Pax8's Q2 2026 survey of 402 US small business leaders found 61 percent actively using AI and another 29 percent still experimenting. Goldman Sachs found 76 percent of 1,256 US owners using AI in early 2026, with 14 percent saying it was fully embedded in core operations.

Do I need an AI consultant or IT partner to move past testing?

Not as the first step. Agree what AI is for, write a short rule and name who decides, because no outside partner can settle those for you. Bring in help first only when the gap is technical, such as security settings you cannot configure or systems holding customer records that need connecting.

What is the Readiness Test?

The Readiness Test is three questions an organization must be able to answer consistently, from the top to the front line, before it can call itself AI ready: What is AI for here? What may I do with it? Who decides when it goes wrong? It is Dan Fitzpatrick's suggested check, not a published standard.

How long should a small business spend testing AI?

There is no researched number, so set your own date. Ask three people what AI is for in your business, fix whatever differs in their answers, then ask again after 30 days. If the answers match and one job now runs on AI by default, the testing is over. If not, repeat the week.

Dan Fitzpatrick helps businesses use AI well through keynotes, practical AI training for teams, and AI strategy and governance for leadership teams.

Dan Fitzpatrick

Written by

Dan Fitzpatrick

Dan Fitzpatrick is The AI Educator: a Forbes contributor, international keynote speaker and author who has trained more than 150,000 people across 30+ countries. He works with businesses and leadership teams on AI training, AI strategy and governance, and keynotes.

He also works with schools as an AI keynote speaker for schools, through school AI strategy consulting and with AI training for schools.