What the surveys say about AI use and training
Plenty of surveyed owners already use AI. Fewer have it built into how the work actually gets done. And a lot of them say they want training.
In October 2026, Goldman Sachs announced $500 million for the next phase of its 10,000 Small Businesses program, according to Fortune and The Next Web. The program covers education, access to capital and support services. The announcement also included a Small Business AI Institute at Goldman's lower Manhattan headquarters, with Anthropic, Google, Meta and OpenAI as partners. It shows training is getting attention. What works is a different question.
A Babson College survey of Goldman Sachs 10,000 Small Businesses alumni, reported by Fortune, found that 86% use AI, while 15% say it's fully integrated into core operations. The survey ran in August 2026 and drew 2,473 responses, according to The Next Web, which also reports that Goldman says respondents may not represent all alumni. Fortune also reports that 77% of them wanted more training and resources. In a March 2026 Goldman survey of its program participants, 76% used AI, 14% were fully integrating it, and 73% said more training and resources would help.
A Thryv survey of 561 small and mid-sized owners and decision-makers lands in the same place: 86% were at least somewhat comfortable with AI, yet about 70% said they needed more training. Thryv sells AI-enabled marketing software and sponsored the survey. As Thryv's president put it, "being comfortable isn't the same as being effective."
None of these is a random sample of small businesses. Goldman's and Babson's respondents are program participants and alumni, so the numbers show a pattern in those groups. They aren't a count of every small business in the country.
Where owners learn today
In that same Thryv survey, which the vendor sponsored, 57% of owners said their main training source is YouTube and social media, and 49% rely on online resources and webinars. These surveys count who wants training. None of these surveys measures whether training changes results.
Other programs exist too. Google and the U.S. Chamber of Commerce launched Small Business B(AI)sics in October 2025, with local workshops and online courses. America's SBDC describes a Google AI Professional Certificate for business owners at no cost. Look at those before you pay anyone.
A first-month plan
This plan is TGC's own. It follows advice that vendors, consultants and trade bodies tend to give, and no source tests the order. It also lines up with what the U.S. Chamber of Commerce reports about early adopters: they start with small tests on real day-to-day problems, invest in training, and keep a person in the loop.
Week 1: write one rule. Keep it to one page. Which AI tools are approved, what data never goes in (customer and financial data, to start), and who checks the output. Start here, because 50% of AI users in Goldman's survey named data privacy and security as a challenge. Tell people what the tools are for, too. In the Babson survey, 84% of respondents reported no change in staffing, 5% added positions and 6% cut them. That's one group of alumni, not a promise about your business.
Week 2: pick one team and one real task. Skip "AI in general." Choose a job people already do every week, like writing customer replies or summarizing call notes. Set a measure before you start, like time per task or how often the output needs fixing. The owner or a team lead should use the tools first, out in the open, where everyone can see.
Week 3: do the task with AI, on real work. Name one person who checks the output, and have staff report what they had to correct. Those corrections show you what to fix. Put the training time on the calendar too, because a short-staffed team cuts it first.
Week 4: recheck the measure. Compare it to where you began. Then make the call: keep going, change the approach, or stop. Check again at about 60 and 90 days.
Signs that training is not enough
Training covers what people can do in a chat window. Some work needs more than that. These signs are TGC's reading of the pattern, not a measured result:
- Several people repeat the same task.
- The output needs the same fix every time.
- The work touches a system staff can't reach from a chat window.
- The process needs to run without a person starting it.
The data fits that pattern, loosely. Use is broad and the depth is thin: in the Babson survey, 18% of respondents use agentic AI, meaning AI that can carry out steps on its own. When a task shows any of these signs, it's worth asking whether to build it. A separate post, on whether AI automation is worth it, covers that decision. This post is about teaching staff first.
Where TGC fits
TGC offers free AI workshops and sells one-on-one AI training and builds, so weigh this advice knowing that. The workshops run in person or online. The AI Classes page has the details, and a build comes after, only when a task has earned it. To talk through your team's first month, book a call with Jason.
