Most Small Businesses Use AI Now. Few Have Built It In. Here Is How to Pick the First Job.
A new Goldman Sachs survey found 86% of its small business owners use AI, and only about one in seven have made it part of how the work gets done. The gap is a choice of job, not a choice of model.
By John Martines · October 31, 2026 · 6 min
Goldman Sachs just opened a Small Business AI Institute, and its own survey shows the gap: almost everyone uses AI, few have built it into the work. Here is how to choose the one job to build in first, and how to know it worked.
The owner of a twelve-person roofing company uses AI every week. He asks it to clean up a customer email, to explain a permit rule, once to write a job posting. Each time it helps. And each Monday, his office still builds the week's estimates exactly the way it did five years ago, by hand, from the same three spreadsheets.
That is the most common place a small business sits with AI right now. It's in use. It isn't in the work.
Last week one of the largest small business programs in the country put a number on that gap, and then put money behind closing it.
On October 22, 2026, Goldman Sachs opened a Small Business AI Institute at its headquarters in lower Manhattan, as part of a new $500 million commitment to its 10,000 Small Businesses program. Anthropic, Google, Meta and OpenAI are partners. Chief executive David Solomon described three parts: training from each partner, research on how small businesses actually use AI, and access to capital. Google, for one, will offer its Grow with Google AI Professional certificate to almost 20,000 small business owners and employees, with three months of free access to Google AI Pro.
Alongside it came a survey of 2,473 owners who went through the program, run by Babson College in August 2026. The headline numbers:
- 86% use AI in their business, up from 56% in 2024.
- About 15% say AI is fully built into their core operations.
- 93% of those using AI report a positive effect, and 74% point to time saved.
- 84% of those using AI say their headcount hasn't changed because of it; 5% added jobs and 6% cut them.
These are owners who chose a growth program, so they are not every small business. The pattern is still worth taking seriously: nearly everyone has tried it, few have changed how the work is done.
Solomon added a point that got less attention. Most of the economy, he said, doesn't need the newest, most powerful AI models; Goldman itself does very little that requires them. For a small business that is good news. The gap isn't about having the best model. It's about picking a job and building it in.
How it got here:
- 2024: 56% of the program's owners reported using AI.
- March 17, 2026: an earlier Goldman Sachs survey found most owners using AI but only 14% with it fully built into core operations, and most asking for training.
- October 22, 2026: the Small Business AI Institute opened, with the August survey showing use at 86%.
Using AI and building it in are different things.
Using it is a person opening a chat when they remember to. The benefit depends on who remembers, and it walks out the door with them.
Building it in means one job in your business now runs a set way, with AI doing a defined part, a person checking a defined part, and the same result whoever is working that day. The estimate gets drafted from the job notes every time. The follow-up email goes out every time. The new hire does it the same way on day one.
That is where the time comes back. A clever prompt saves one person ten minutes once. A job that runs the same way every week saves the whole office hours, every week, and it keeps working when someone is out sick.
1. Pick one job, not "AI." Look for work that happens every week, follows roughly the same steps, starts from something written (notes, a form, an email) and ends in something written (an estimate, a reply, a schedule). A dental office's appointment reminders and recall letters. A contractor's estimate from site notes. A shop's product descriptions for new stock. If the job happens once a quarter, skip it for now.
2. Write down how it's done today. Before any AI, have the person who does the job walk through it, step by step, while someone types. Where does the information come from, what decisions get made, what does a good result look like? This one page is most of the work. Our piece on [why writing down your processes is the advantage that lasts](/articles/process-capture-is-the-only-ai-moat-left) goes deeper on how.
3. Mark the AI part and the person part. On that page, circle the steps AI can draft (pulling the details together, writing the first version) and box the steps a person always owns (prices, promises, anything sent to a customer). Most first jobs end up as "AI drafts, a person checks and sends."
4. Use the tools you already pay for. You probably don't need a new product. Most small businesses already have ChatGPT, Claude, Copilot or Gemini through an existing plan. Save the instructions for the job in one place, as a project, a custom assistant or simply a document everyone pastes from, so it runs the same way for everyone.
5. Run it for four weeks and count. Note how long the job took before, and how long it takes now, including the checking. Note every mistake the check caught. If it saves time and the mistakes are the kind a person catches easily, it's built in. If not, adjust the instructions or pick a different job.
6. Then pick the next one. One job built in properly teaches the office more than ten people experimenting on their own. After the first, the second goes faster.
- Is it written down, so a new hire could do it from the page?
- Does it run the same way no matter who is working that day?
- Is the person-checks step named, with a name next to it?
- Do you know how long it took before and how long it takes now?
- If the person who set it up left tomorrow, would it keep running?
Training is about to get easier to find. Goldman's institute, its four AI partners and offers like Google's certificate will put more courses in front of owners this winter. Training helps, and most of the owners surveyed said more of it would. But a course teaches what the tools can do. Only you can decide which job in your business they should do every week.
Pick the job this week. Write down how it's done today. Then build AI into that one job, check it, count the time, and move to the next. That is how a business goes from using AI to running on it.
---
Sources (checked October 28, 2026). Small Business AI Institute, the $500 million commitment, partners, its three parts, Google's certificate offer and David Solomon's remarks on frontier models: Fortune, report by Nick Lichtenberg on the institute launch (Oct 23, 2026). Launch date and partners: Crypto Briefing, "Goldman Sachs launches $500 million AI push for small businesses" (Oct 22, 2026). August 2026 survey of 2,473 program alumni by Babson College, including 86% use (56% in 2024), about 15% fully integrated, 93% positive effect, 74% time saved, workforce effects and the call for more training: Fortune (Oct 23, 2026), NewKerala/ANI (Oct 26, 2026) and AOL (Oct 2026); reported counts differ by one respondent across outlets. Earlier survey with 14% fully integrated: "Survey: Small Businesses Embrace AI But Need Training and Support to Fully Harness It" (Goldman Sachs via Business Wire, Mar 17, 2026).