86% use AI. Only 15% have it built in. What Goldman's new survey means for your office

What happened

On Oct. 22, 2026, Goldman Sachs announced a Small Business AI Institute inside its 10,000 Small Businesses program. Anthropic, Google, Meta and OpenAI are partners. Google will offer its AI Professional certificate to almost 20,000 owners and employees. Anthropic will build a learning series on applied AI, and OpenAI will run workshops with the program's community college partners.

The institute is part of a new $500 million commitment to the next chapter of the program, which also covers education, access to capital and support for another 10,000 businesses. So it isn't a $500 million AI fund. The more useful part for an owner is the survey that came with it.

What the survey found

Babson College ran the survey from Aug. 10 to 24, 2026. It went to 14,967 program alumni and got 2,473 responses, a 16.5% response rate. These are owners who went through a Goldman program, so read the numbers as a picture of that group.

AI use vs. AI built into operations

Goldman Sachs 10,000 Small Businesses alumni survey, Aug. 2026 (2,473 responses). Self-reported. "AI users" figures are shares of owners who use AI.

1. Use AI in the businessAll respondents86%
2. No formal employee AI policyAI users74%
3. Use AI for financial analysis and planningAI users42%
4. Use agentic AIAI users18%
5. AI fully built into core operationsAll respondents15%

Source: Goldman Sachs, The AI Economy Is a Small Business Economy, Oct. 2026.

A few other findings from the report:

• Jobs mostly unchanged. Among owners using AI, 84% report no workforce change because of it, 5% added positions and 6% cut staff.
• Cheap tools. 73% of AI users spend less than $100 a month on AI tools, including those on free tools only.
• Bigger firms, less employee use. 28% of businesses with 1 to 25 employees say all employees use AI. For businesses with 26 or more, it's 11%.
• Product businesses lag service firms. 80% of product businesses use AI, against 87% of service businesses. 14% have it fully integrated, against 19%. But product businesses use it more for purchasing (22% vs. 10%) and for distribution and logistics (15% vs. 6%).
• Why non-users hold back. Among the owners not using AI, 45% don't know how to apply it, 32% cite data privacy and 28% worry about inaccurate results.

Most offices use AI tools. Few have changed how the work actually flows.

Why the 86% vs. 15% gap matters to you

Using AI usually means someone asks a chatbot to draft an email or summarize a document. That's useful. But the order still gets typed into the ERP by hand. The invoice still gets matched by eye. The work moves at the same speed, with a bit of help on the side.

Built in means the process itself changed. The PO is read and drafted into the ERP, and a person approves it. The quote pulls last year's job automatically. That's the 15%.

The size finding makes sense. In a 10-person firm, the owner tries a tool and everyone picks it up. In a 150-person company, use is uneven. Purchasing tries one thing, the order desk another, and nobody is sure what's allowed. That's how you end up in the 74% with no policy.

The product-business numbers are worth a second look if you make or distribute things. Product firms in the survey were slower to adopt and less likely to have AI built in. Yet they were twice as likely as service firms to use it for purchasing, and more than twice as likely for distribution and logistics. That's where the paperwork is: purchase orders, confirmations, ship notices, invoices. It's also where a rebuilt process pays back fastest, because the same steps repeat hundreds of times a month.

Where the skeptics have a point

• It's not a national sample. These are alumni of a Goldman program who chose to answer, about one in six. Owners who like AI may have been more likely to reply. As one newsletter put it, the figures describe Goldman alumni, not small businesses nationwide.
• Everyone here has something to sell. The four partners sell AI. And NBC News reports that Goldman is reportedly working on Anthropic's IPO.
• Caution is reasonable. NBC spoke with the head of Linda Tool, a Brooklyn maker of precision parts for defense and aerospace. He is careful about AI because of the sensitive data his company handles. He also wants AI to support his people, not replace them. "Machines are expensive," he told NBC. "People are priceless."

None of that makes the gap less real. It just means you should measure your own office, not lean on someone else's average.

What to do this quarter

1. Find out who uses what. Ask each department which AI tools they use and for what. Expect more than you think.
2. Write a one-page policy. Which tools are approved, what data can't go into them (customer pricing, employee records, anything under NDA), and who to ask. If you're one of the 74%, this is the cheapest fix on the list.
3. Pick one process, not one tool. Choose a repeated job with a clear input and output: purchase orders in, invoices to match, quote requests. Measure minutes per item today.
4. Build it in, with a person approving. Software does the reading and typing inside the process. Your team checks and approves. Then measure again.
5. Use the free training. If you qualify for the program's courses, send the people who'll run the process, not just the owner.

The survey says most owners have already tried AI. The next step isn't another tool. It's one process that actually runs differently on Monday.