AI-generated article. This article was researched and drafted using AI tools and published automatically, and its featured image was generated by AI. Facts are drawn from the sources cited in the text.
Most owners reading this already have a chatbot open in another tab. They use it for emails, for pricing questions, for the first draft of a proposal. And yet the business itself runs more or less the way it ran two years ago. That quiet distance between using the tools and actually changing how the work gets done now has hard numbers behind it, and it is the clearest description yet of the small business AI skills gap.
Adoption is finished. Integration barely started.
Goldman Sachs published its 2026 alumni survey on October 22, run by Babson College across 2,473 responses from graduates of its 10,000 Small Businesses program. Eighty-six percent use AI in the business, up from 77% in 2025 and 56% in 2024. Only 15% say it is fully integrated into core operations. Half the users are active in some areas, another 32% just experimenting.
Those three numbers sit uncomfortably together. Near-universal use. Almost no embedding. A busy middle that is not systematic. The same pattern shows up whenever a tool gets cheap before anyone works out how to move it from pilot into production.
The small business AI skills gap is really a distribution problem
Look at who inside the company actually touches the tools and the picture sharpens. In 14% of AI-using businesses the owner is the only person using it. In 42%, some employees do. Just 26% say everyone does. And 74% of owners using AI have no formal policy for staff use at all.
That is the small business AI skills gap in one line. The knowledge lives in one head, usually the busiest head in the building, and nothing around it is written down.
Smaller firms do better here than you might expect. Among businesses with 1 to 25 employees, 28% have every employee using AI, against 11% of firms with 26 or more. Being small is an advantage for spreading a habit. It rarely feels like one.
How representative is 86%?
Worth holding lightly. Goldman’s respondents graduated from an intensive business program, about as self-selected as a sample gets, and the report says so. Broader data lands lower. The Federal Reserve Bank of New York, writing in October on the 2025 Small Business Credit Survey of 5,248 employer firms, put AI use at 46%, with 51% of those users having partially or fully integrated it. The JPMorganChase Institute, tracking payments across 4.6 million Chase business accounts, found 17.7% were paying for an AI service by the end of 2025.
Different questions, different answers. The direction is identical. Use is climbing faster than depth.
Half a billion dollars, aimed at training rather than software
Goldman committed $500 million to the next phase of the program and launched a Small Business AI Institute with it, with Anthropic, Google, Meta and OpenAI each supplying education rather than discounts. Google is putting its AI certificate and three months of AI Pro access in front of nearly 20,000 owners and employees. OpenAI is running workshops through community college partners.
What they chose to fund is the interesting part. Seventy-seven percent of surveyed graduates, users and non-users alike, said they would benefit from more training. Among the 14% not using AI at all, the top reason was not cost or fear. It was not knowing how to apply it, cited by 45%.
David Solomon, Goldman’s chief executive, told Fortune that for small business work there is very little that needs frontier models, which are separate from the base tools that already exist. Unusually plain, for a bank chief in 2026. The constraint is not model quality. It is what people know how to do with what they already pay for, and 73% of these owners pay under $100 a month. A TD Bank survey reported by Forbes in October found 97% of owners feel unprepared in at least one key area, with 24% naming a lack of employee training outright.
One thing that would move the number
Pick one process that repeats every week. Invoice chasing, intake emails, the Monday numbers, the social post. Write down, in plain language, the steps someone would follow to do it with the tool. Then hand that page to one other person and let them run it twice.
That is it. Not a strategy, not a platform. A written procedure and a second person who can run it. It is the unglamorous middle of AI training for a small business, and it is where judgment about the output gets built, because the second person finds the places where the tool is wrong.
It attacks exactly what the data says is missing: capability stuck with one person. Ninety-three percent of AI users in the Goldman survey say the impact has been positive, while 84% report no workforce change at all. The tools are helping. They are helping one person.
The overwhelm is real too. Most owners are not avoiding integration out of reluctance. They are running a business with no spare afternoon in it, while every vendor shouts about agents and the honest next step is smaller and far less exciting. Every little tweak still counts. Anyone who writes down one process this month is already on a good path.
Which leaves a question the $500 million does not answer. If the real constraint is time and written process rather than literacy, does more training close the gap, or does it just add to the pile of things an owner already knows but has not yet had a week to build?
