The AI Skills Gap: What Small Businesses Are Actually Missing

The AI skills gap is the distance between what a business needs its people to know to get value from AI and what they know today. For a company of 10 to 200 people, the missing skills are rarely the ones in the headlines. They're operating skills, like picking the right piece of work, writing down what a good result looks like, checking what the AI produces, and owning the workflow after launch, and they sit closer to your best project manager or office manager than to a programmer, which is why hiring a specialist rarely closes the gap on its own.

The gap is real and it is large. In the European Union's 2025 survey of business technology use, 70.3% of enterprises that had considered AI but weren't using it named a lack of relevant expertise as a reason, ahead of legal uncertainty, privacy concerns, data problems, and cost. Missing skills are also behind many stalled AI projects, and our guide to why AI projects fail covers the others, from choosing the wrong problem to having no one who owns adoption.

Key Takeaways

• The AI skills gap that stalls projects at small and mid-sized businesses is mostly a gap in operating skills, not a shortage of data scientists or machine learning engineers.
• In Eurostat's 2025 survey, 70.3% of EU enterprises that had considered AI but weren't using it cited a lack of relevant expertise, the most common reason given. Small firms (70.9%) cited it more often than large ones (65.1%).
• Businesses mostly plan to train, not hire. Among Canadian businesses planning to use AI, 49.8% expected to train current staff and only 12.6% expected to hire staff trained in AI.
• Training lags demand. In Microsoft and LinkedIn's 2024 survey, 66% of leaders said they wouldn't hire someone without AI skills, while only 39% of AI users had received AI training from their company.
• Judgment is the skill that protects results. In a 758-person experiment at BCG, consultants using AI did more work, faster and better, on tasks inside the AI's abilities, and were "19% less likely to produce correct solutions" on a task outside them.
• Your experienced people are the raw material. In a study of 5,179 support agents, an AI assistant raised productivity 14% on average and 34% for novices, with suggestive evidence that it passed on the practices of the best workers.

Table of Contents

• What is the AI skills gap?
• How big is the AI skills gap for small businesses?
• Why doesn't hiring an AI specialist close the gap?
• The five skills AI projects actually run short of
• Why your most experienced people are the skill
• A 65-person landscaping company closes its gap
• How to close the AI skills gap without hiring
• Where Avolis fits
• Frequently Asked Questions
• Continue Learning

What Is the AI Skills Gap?

The AI skills gap is the shortfall between the AI-related skills a business needs and the skills its workforce has. Most published discussions define it as a shortage of technical talent, meaning people who can build, train, and deploy models. That definition fits a bank or a software company building its own AI. It fits much less well for a business that uses AI tools someone else built, which describes most companies with 10 to 200 employees.

For that second kind of business, the work that decides whether AI pays off happens before and after the technology. Someone has to choose a task worth handing to AI, describe the standard the output has to meet, catch the cases where the AI is confidently wrong, and keep the workflow running when prices, staff, or software change. A project can have a capable tool and a competent vendor and still fail because nobody inside the business can do those four things.

Here's the plain version. A small business needs very little skill at making AI, and a good deal of skill at directing and checking it, and that second kind of skill is the one most training budgets and hiring plans skip.

How Big Is the AI Skills Gap for Small Businesses?

Among businesses that have considered AI and aren't using it, a lack of expertise is the most common reason given, and small firms report it more often than large ones. Eurostat's 2025 survey of enterprises with 10 or more employees asked firms that had considered AI, but weren't using it, why not. A lack of relevant expertise topped the list at 70.3%, well ahead of unclear legal consequences (53.6%), data protection concerns (52.7%), problems with data (43.5%), and cost (38.4%) (Eurostat, dataset isoc_eb_ai, updated June 2026).

The size pattern matters for a smaller operation. In the same survey, 70.9% of small firms (10 to 49 employees) cited a lack of expertise, against 69.2% of medium firms and 65.1% of firms with 250 or more. Small firms also used AI the least: 17% of them used at least one AI technology in 2025, against 30% of medium firms and 55% of large ones (Eurostat, Use of artificial intelligence in enterprises, Statistics Explained, December 2025). The firms with the fewest specialists to call on cite expertise most often, and use AI least.

North American data tells a similar story with a twist. In Statistics Canada's survey from the third quarter of 2025, two-thirds of businesses had no plans to use AI in the next year, and 78.1% of those said AI wasn't relevant to the goods or services they provide. Only 4.6% blamed a lack of skilled workers, and 11.3% cited a lack of knowledge about what AI can do (Statistics Canada, Analysis on expected use of artificial intelligence by businesses in Canada, third quarter of 2025, September 11, 2025). Deciding that AI has nothing to offer a roofing company or a property manager is itself a judgment, and it's a hard one to make well without first counting where the office hours go. A missing skill can show up as "not relevant" as easily as it shows up as "we don't have the expertise."

Training hasn't kept up either. When Microsoft and LinkedIn surveyed 31,000 people in 31 countries in 2024, two-thirds of leaders (66%) said they wouldn't hire someone without AI skills, yet only 39% of AI users had received AI training from their company, and only 25% of companies expected to offer it that year (Microsoft and LinkedIn, 2024 Work Trend Index, May 8, 2024).

Why Doesn't Hiring an AI Specialist Close the Gap?

Hiring an AI specialist rarely closes the skills gap at a small business, because the missing skills belong to the people who run the work, and a new hire doesn't know your work. A data scientist can build a model, but a 60-person electrical contractor usually doesn't need a model built. It needs someone who knows which bids lose money, which customer emails are urgent, and what a clean closeout package contains, and that person is already on the payroll.

Businesses' own plans point the same way. Among Canadian businesses planning to use AI in the next 12 months, training current staff was the most expected change, at 49.8%, while hiring staff trained in AI was the least expected, at 12.6%. Businesses already using AI reported the same pattern looking back: in the previous year, 40.1% had developed new workflows and 38.9% had trained current staff, while hiring staff trained in AI "remained relatively uncommon" (Statistics Canada, 2025).

Cost is part of it: the median data scientist in the United States earned $120,230 in May 2025 (U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Data Scientists), and a small business competes for that pool against banks and software companies. Our comparison of AI development companies and in-house teams works through the full cost of an internal hire.

The bigger reason is fit. U.S. Census Bureau researchers found that 18% of American firms used AI in a business function in late 2025 and early 2026. Of those users, 57% had AI in three or fewer functions, and 66% used it only to help people with tasks rather than to take tasks over (Bonney et al., The Microstructure of AI Diffusion, Census Bureau Center for Economic Studies Working Paper 26-25, April 2026). That's the typical shape of AI in a smaller firm: a few workflows, each run by the people who already do the job, with AI handling a draft or a first pass. A full-time specialist in that setting ends up with a thin slice of technical work and a large amount of work nobody hired them for, which is learning how the business runs.

The Five Skills AI Projects Actually Run Short Of

AI projects at small and mid-sized businesses run short of five skills: choosing the work, defining good output, checking the output, owning the workflow after launch, and connecting AI to the systems the business already uses. Only the last one is technical, and it's the easiest to rent. The other four have to live inside the business, because they depend on knowing how the work is actually done.

1. Choosing the work

The first skill is picking a task where AI's strengths match a real cost in the business. That takes two kinds of knowledge at once: a rough sense of what current AI does well (reading and drafting documents, summarizing, sorting, extracting fields from forms) and a precise sense of where your hours go. Owners usually have the second and lack the first, and vendors have the first and lack the second.

When this skill is missing, the project goes to whatever the tool demonstrates best, or to the task the owner finds most annoying, which isn't always the task that costs the most. The most useful exercise is unglamorous: count how often a task happens, how long it takes, and how many hands it passes through, then rank. Our page on the current wave of AI project failures shows what happens when starting is cheap and this step gets skipped.

2. Defining what good output looks like

The second skill is writing down the standard the AI's work has to meet, in terms specific enough that someone else could check it. Most businesses have never needed to do this, because the standard lives in an experienced person's head. A senior project manager at a cabinet and millwork shop knows that an install on a third-floor walk-up gets a labor bump, and that one hardware supplier's lead times run two weeks longer than quoted, and nobody ever wrote either rule down.

AI doesn't know those rules either. Without them, its drafts look complete and miss the things that matter, and the people receiving them start rewriting every one. Defining good output usually means a short list (eight to fifteen rules is common) that names the checks, the exceptions, and the cases that should go to a person instead of getting a draft at all.

3. Checking the output

The third skill is knowing where AI is likely to be wrong and catching it before it reaches a customer. The best evidence on why this matters comes from a preregistered experiment with 758 consultants at Boston Consulting Group. On 18 realistic tasks within the AI's capabilities, consultants using GPT-4 completed 12.2% more tasks, 25.1% faster, with significantly higher quality. On one task deliberately chosen to sit outside those capabilities, consultants using AI were "19% less likely to produce correct solutions" than those working without it, in the words of the paper's abstract (Dell'Acqua et al., "Navigating the Jagged Technological Frontier", Harvard Business School Working Paper 24-013, 2023; published in Organization Science, 2026).

The researchers called this boundary a "jagged frontier," because tasks of similar difficulty can fall on either side of it. The practical skill is knowing which side a given task is on, and the people best placed to learn that are the ones who already do the work and know what a wrong answer looks like. Checking is a skill people build, through practice on real cases where they can see the AI's misses. A general course on writing prompts doesn't teach it.

4. Owning the workflow after launch

The fourth skill is keeping an AI workflow accurate and in use as the business changes around it. Prices change, a new service line starts, a key employee leaves, and the software the AI reads from gets an update. Each of those can quietly break a workflow that worked on launch day.

Ownership is less a technical skill than a job assignment: one named person who reviews a sample of the AI's work on a schedule, updates the rules when the business changes, and decides when something needs to be fixed by whoever built it. A pattern we often see in diagnostics is a stalled project that had a builder and users, and nobody in between. Without that owner, there's also nobody to say whether the workflow is paying for itself, which is the question our guide to how to measure AI ROI answers.

5. Connecting AI to your systems

The fifth skill is the technical one: getting AI to read from and write to the systems the business runs on, such as the job management system, the accounting software, a document folder, or the email inbox. This is where the conventional picture of the skills gap is accurate, since it takes real integration experience to do reliably.

It's also the skill a small business most sensibly rents. Statistics Canada found that 16.0% of businesses planning to use AI expected to use vendors or consulting services to install or integrate it, a larger share than expected to hire. The skill a business can't rent is knowing whether the connected system is doing the right thing, which brings the work back to the first four skills.

Why Your Most Experienced People Are the Skill

The strongest evidence that the AI skills gap is mostly about operating knowledge comes from a study of AI in a real workplace. Researchers followed the staggered rollout of an AI assistant to 5,179 customer support agents at a software company. Access to the tool raised productivity, measured as issues resolved per hour, by 14% on average and by 34% for novice and low-skilled workers, with minimal effect on the most experienced and skilled agents. The researchers report suggestive evidence that the AI was spreading the practices of the most able workers to newer ones (Brynjolfsson, Li, and Raymond, "Generative AI at Work", NBER Working Paper 31161, revised November 2023; published in the Quarterly Journal of Economics, 2025).

That finding changes how a small business should think about its skills gap. The assistant in that study was trained on the conversations of the company's best agents, so the value it delivered to new hires was, in effect, the know-how of its veterans in a form that could be passed along. A smaller business can't train its own model, but it gets the same effect when its best people write down the rules, review the drafts, and correct the misses, because their judgment is what turns a general-purpose tool into one that works for this business.

The trade-off is time. The senior people whose knowledge the AI needs are usually the busiest people in the company, and pulling them off billable work to write rules and review drafts feels expensive in the first month. It's also the part of the project that holds its value after the vendor leaves, since the rules and the review habit stay with the business.

A 65-Person Landscaping Company Closes Its Gap

Consider a 65-person commercial landscaping and snow-removal company serving office parks, homeowner associations, and retail centers (an illustrative composite). Its account managers write proposals for "enhancement" work beyond the maintenance contract: replacing dead shrubs, repairing irrigation, adding seasonal color. Each proposal took a site walk, a price build in a spreadsheet, and a write-up, and the backlog of unsent proposals regularly ran past 40.

The owner's first move was to hire a freelance developer to build a proposal generator. It took site-walk notes and photos and produced a priced, formatted proposal in a few minutes. Two months in, the two newest account managers were using it and the three senior ones weren't. The senior managers had found that it priced irrigation repairs at the published rate without the company's two-hour minimum, ignored the surcharge on properties that needed a lift or after-hours work, and quoted plant material at prices that were a season out of date. One underpriced proposal for a retail center went out signed, and the company lost money on the job. After that, every generated proposal got rebuilt by hand, and the backlog grew.

Every missing piece was an operating skill. The pricing rules lived in the heads of the senior managers, so nobody had defined good output. The newer managers didn't know which numbers to distrust, so nobody could check it. When the nursery raised its prices in spring, nobody was responsible for updating the tool, so nobody owned it.

The second attempt started with the company's longest-serving account manager, 18 years in the trade. She wrote 11 pricing rules covering minimums, access surcharges, crew-size thresholds, and the cases that always need a site re-check, and she became the workflow's owner, with Tuesday mornings blocked out for the job. The office manager took over the plant-material price list and updated it monthly from the nursery's sheet. Each new account manager spent their first two weeks having every generated proposal reviewed against the rules, with the misses discussed at a weekly 30-minute meeting. The developer, kept on a small monthly retainer, connected the generator to the price list and the company's job management system.

After a full season, all five account managers used the generator for every enhancement proposal, the median time from site walk to sent proposal had fallen from six business days to two, and the backlog of unsent proposals had dropped below ten. The newest account managers' proposals were being accepted at close to the rate of the senior managers', the same mechanism the support-agent study describes, with the senior people's judgment, captured in rules and reviews, doing the most for the people with the least experience.

How to Close the AI Skills Gap Without Hiring

A business of 10 to 200 people can close most of its AI skills gap with the people it has, by assigning each of the four operating skills to a named person and renting the technical one. Each skill has a natural owner, a way to build it, and a sign that it's working.

Two habits make the plan hold. Have people practice on last month's actual quotes, letters, or work orders rather than on generic examples, because checking is a skill that depends on knowing what a wrong answer looks like in your business. And give the owner real hours in the week, because ownership that has to fit around a full workload is the first thing to slip. Our page on corporate AI implementation failure covers the rest of the people side of a rollout, from what to say about jobs to where the saved time should go.

Citation-ready summary: The AI skills gap at small and mid-sized businesses is mainly a gap in operating skills, not technical talent. In Eurostat's 2025 survey, 70.3% of EU enterprises that had considered AI but weren't using it cited a lack of relevant expertise, and small firms cited it most. Businesses close the gap mostly through training: 49.8% of Canadian businesses planning to use AI expected to train current staff and 12.6% to hire AI-trained staff. The skills that matter most are choosing the work, defining good output, checking the output, and owning the workflow, with system integration the one skill sensibly rented.

Where Avolis Fits

Every engagement starts with our diagnostic, which does the first of the five skills with your team: we work through each candidate workflow with the people who run it, measure volumes, time per task, and handoffs, and rank where AI would actually pay off. You keep that ranked result whatever you decide next. We also map who in the business already holds the judgment each workflow needs, since that person usually becomes its owner.

When we build, we handle the integration, and we work with your experienced people to write the rules the AI has to follow and the checks your team runs on its output. Because we stay embedded with your team after launch, we help the workflow owner through the first price change, the first staff change, and the first software update, so the skills stay in your business rather than leaving with the project. Our guide to AI readiness assessment services explains what a good diagnostic should produce, and our guide to which AI consulting company to choose covers how to tell a partner who will build these skills with you from one who hands over a tool.

Frequently Asked Questions

What is the AI skills gap?

The AI skills gap is the shortfall between the AI-related skills a business needs and the skills its people have. At large technology companies it usually means a shortage of engineers and data scientists. At smaller businesses it mostly means missing operating skills: choosing the right task, defining good output, checking what the AI produces, and owning the workflow after launch.

How common is the AI skills gap?

It's the most common reason given by businesses that have considered AI and aren't using it. In Eurostat's 2025 survey, 70.3% of EU enterprises that had considered AI but weren't using it cited a lack of relevant expertise, ahead of legal, privacy, data, and cost concerns. Small firms with 10 to 49 employees cited it more often (70.9%) than firms with 250 or more (65.1%).

Should a small business hire an AI specialist to close the skills gap?

Usually not as a first step. The skills most small-business AI projects lack belong to the people who already run the work, and a specialist would have to learn the business first. Most businesses train instead: in Statistics Canada's 2025 survey, 49.8% of businesses planning to use AI expected to train current staff, and only 12.6% expected to hire AI-trained staff.

What AI skills do employees actually need?

Most employees need to know where AI tends to be wrong in their own work and how to check its output against the business's rules. A BCG experiment with 758 consultants found that AI made them faster and better on tasks within its abilities, and "19% less likely to produce correct solutions" on a task outside them. Knowing which kind of task is in front of you is the core skill.

How do you close the AI skills gap without hiring?

Assign each operating skill to a named person. Have the owner and team managers rank tasks by measured hours, have your most experienced person write the rules a good result must follow, train users on reviewed practice with their own recent work, and give one person regular hours to own the workflow. Rent the system integration from a vendor or consultant.

Continue Learning

The AI skills gap is real, and for a smaller business it's mostly closable from the inside, because the scarce skills are the judgment your experienced people already have and the habit of writing it down. A tool and a vendor can supply the technical piece. The business supplies the standard, the checking, and the owner, and projects that skip those tend to stall no matter how capable the tool is.

If you're planning a first AI project, name the person who will own it and the person whose rules it will follow before you choose any software.

Why projects fail, in detail:

• Why do AI projects fail? Common reasons and statistics
• Why are AI projects failing?
• Corporate AI implementation failure: why rollouts stall after the tool works
• AI project failure statistics
• How to measure AI ROI

Starting the right way:

• AI readiness assessment services
• Which AI consulting company should I choose?
• AI development companies vs. an in-house team

---

Sources

• Eurostat, "Artificial intelligence by size class of enterprise" (dataset isoc_eb_ai), indicators E_AI_BLE, E_AI_BLEG, E_AI_BCDP, E_AI_BDDT, E_AI_BINC, E_AI_BCST, E_AI_BNU, unit "percentage of the enterprises which ever considered to use any of the AI technologies", EU27, 2025, updated 15 June 2026, retrieved 2026-10-26 via the Eurostat dissemination API: https://ec.europa.eu/eurostat/databrowser/view/isoc_eb_ai/default/table
• Eurostat, "Use of artificial intelligence in enterprises," Statistics Explained, data extracted December 2025, retrieved 2026-10-26: https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Use_of_artificial_intelligence_in_enterprises
• Valerie Bryan, Shivani Sood, and Chris Johnston, "Analysis on expected use of artificial intelligence by businesses in Canada, third quarter of 2025," Statistics Canada, Analysis in Brief 11-621-M, 11 September 2025, retrieved 2026-10-26: https://www150.statcan.gc.ca/n1/pub/11-621-m/11-621-m2025011-eng.htm
• Microsoft and LinkedIn, "Microsoft and LinkedIn release the 2024 Work Trend Index on the state of AI at work," 8 May 2024, retrieved 2026-10-26: https://news.microsoft.com/source/2024/05/08/microsoft-and-linkedin-release-the-2024-work-trend-index-on-the-state-of-ai-at-work/
• U.S. Bureau of Labor Statistics, Occupational Outlook Handbook, "Data Scientists" (May 2025 wage data; 2025–35 projections), retrieved 2026-10-26: https://www.bls.gov/ooh/math/data-scientists.htm
• Kathryn Bonney, Cory Breaux, Emin Dinlersoz, Lucia Foster, John Haltiwanger, and Aditya Pande, "The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks," U.S. Census Bureau Center for Economic Studies Working Paper CES-WP-26-25, April 2026, retrieved 2026-10-26: https://www.census.gov/library/working-papers/2026/adrm/CES-WP-26-25.html
• Fabrizio Dell'Acqua, Edward McFowland III, Ethan Mollick, Hila Lifshitz-Assaf, Katherine C. Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, and Karim R. Lakhani, "Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality," Harvard Business School Working Paper 24-013, September 2023; Organization Science, March–April 2026, retrieved 2026-10-26: https://www.hbs.edu/faculty/Pages/item.aspx?num=64700
• Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, "Generative AI at Work," NBER Working Paper 31161, April 2023, revised November 2023, doi:10.3386/w31161; published in the Quarterly Journal of Economics 140(2), 2025, pp. 889–942, retrieved 2026-10-26: https://www.nber.org/papers/w31161

On the Eurostat figures. The reasons question is asked of enterprises that had considered using AI but weren't using it, and enterprises could name more than one reason. The 70.3% figure is for all enterprises with 10 or more employees; the size-class figures are 70.9% (10 to 49), 69.2% (50 to 249), and 65.1% (250 or more). Statistics Explained shows 70.9% beside 2025 for the overall figure, but the dataset gives 70.89% for 2024 and 70.31% for 2025, so this page uses the dataset value. The AI-use shares by size (17.0%, 30.4%, 55.0%) are from Statistics Explained.

On the Statistics Canada figures. The Canadian Survey on Business Conditions sampled 21,406 employer businesses from 2 July to 6 August 2025, with results weighted to represent all employer businesses in Canada. The 49.8% and 12.6% figures are shares of businesses planning to use AI in the next 12 months (14.5% of all businesses). The 40.1% and 38.9% figures are shares of businesses that used AI in the previous 12 months (12.2% of all businesses). The 78.1%, 11.3%, and 4.6% figures are shares of businesses with no plans to use AI.

On the Census figures. The figures come from the 2026 AI supplement to the Business Trends and Outlook Survey, with a reference period of November 2025 to January 2026. The 18% is unweighted by employment; the employment-weighted share is 32%.

On the two experiments. The BCG experiment's 19% figure is from the Harvard Business School working paper abstract and describes the one task designed to fall outside the AI's capabilities. The support-agent figures (5,179 agents, 14%, 34%) are from the abstract of the November 2023 NBER revision, and this page cites that version throughout.

On the landscaping company. It's an illustrative composite, not a client. Its figures (65 people, a backlog past 40, five account managers, 11 rules, three hours a week, a two-hour minimum, six business days to two, a backlog below ten) are representative examples, not measurements or benchmarks.

---

About Avolis Research Group

Avolis Research Group is Avolis's in-house research practice, focused on how operations-heavy small and mid-sized businesses actually adopt AI. It synthesizes primary economic research, government survey data, and results from real implementations into practical, vendor-neutral guidance.

More about Avolis and how we work · Get in touch