AI & AutomationGuide

86% of Small Businesses in Goldman Sachs' Program Use AI. Only 15% Have Built It Into How They Work.

Using AI and running part of the business on it are not the same thing. Goldman Sachs' new numbers show most small businesses have done the first. Very few have done the second, and the second is where the time and money come back.

Author

DueClix Engineering

Published

Reading time

4 min read

On 22 October Goldman Sachs launched a Small Business AI Institute, with Anthropic, Google, Meta and OpenAI as partners and a new $500 million commitment to its 10,000 Small Businesses program. With it came a survey of more than 2,400 of the program's graduates. Two numbers in it tell the story.

  • 86% said they use AI in their business, up from 56% in 2024.
  • Only 15% said AI is fully integrated into their core operations.
  • More than three-quarters said they could use more training and resources.

So most of these businesses use AI. Very few have made it part of how the business actually runs.

The same gap showed up in March

This is not a one-off. In March, Goldman's 10,000 Small Businesses Voices survey of 1,256 program participants, run by Babson College and David Binder Research, found:

  • 76% were using AI, and 93% of those said it had a positive impact.
  • 84% named efficiency and productivity as the main benefit.
  • Only 14% said AI was fully embedded in their core operations.
  • 73% said more access to training and implementation help would make a difference.
  • The barriers they named: lack of technical expertise, difficulty choosing tools, and data privacy concerns.

Use went up by ten points between the two surveys. Full integration barely moved. They are different groups of people, so the numbers should not be read as a trend line, but they point the same way.

Read them with care

Both surveys are of people who went through Goldman's own program, which includes business education. They are not a random sample of U.S. small businesses, and they are likely to be ahead of the average. The October survey was published alongside an announcement by Goldman and four AI companies, all of whom have an interest in more AI adoption. And "fully integrated" is the respondent's own judgement, not a measured standard.

Even so, the gap between "we use it" and "we run on it" is the most useful part of the data, and it matches what we see.

What "using AI" usually looks like

In most small businesses, using AI means someone opens a chat window to draft an email, summarise a document or write a product description. It helps. But it depends on that person remembering to do it, knowing what to ask, and copying the answer into the right place. When they are off sick, it stops. Nobody checks the output the same way twice, and nobody knows what it costs or whether it is getting better or worse.

Built-in looks different. The quote request arrives, the AI drafts the quote from your price list, a named person checks it against three things, and it goes out from your system. It happens the same way every time, whoever is in the office. That is the 15%.

Why so few get there

The barriers in the March survey are the ones we hear most. Technical expertise: connecting AI to the email inbox, the accounting package or the booking system is real work. Choosing tools: there are thousands, and most demos show the easy 80% of a task, not the awkward 20%. Data privacy: owners are right to ask where customer data goes before they connect anything. None of those are solved by another chat subscription.

Five steps to move one task from "used" to "built in"

  1. Pick one task that repeats every week. Quotes, invoice chasing, appointment reminders, answering the same ten customer questions. Not the most impressive task. The most repetitive one.
  2. Write down how it is done today, step by step. Who starts it, what information they need, where the result goes. If you cannot write it down, AI cannot follow it either.
  3. Decide what the AI does and what a person checks. For example: AI drafts the quote, a person checks the price, the quantity and the delivery date. Write the check down. That is what makes it safe to rely on.
  4. Put it where the work already happens. In the inbox, the CRM or the accounting tool, not in a separate tab someone has to remember. This is usually the step that needs technical help.
  5. Give it an owner and a number. One person who notices when it breaks, and one measure, such as time from request to quote sent. Check it once a month. If the number does not move, change the process or stop.

Then do the next task. Businesses that get to "fully integrated" rarely do it in one project. They do it one workflow at a time. We wrote about why the checking step matters so much in AI was supposed to mean less work, and about automating the wrong thing in the automation trap. For data privacy, our free AI agent safety check covers what to ask before you connect anything.

The short version

Almost every small business in these surveys has tried AI, and most say it helps. Very few have made it part of the way the business runs. That second step is not about a better AI model. It is about picking one task, writing down how it works, deciding who checks it, and putting it where the work already happens.

Sources

Written by

DueClix EngineeringEngineering team

The team that designs and builds DueClix systems. We write about the parts of the work that are worth writing down.

Have a process worth improving?

Let's build the system behind it. Tell us what the process is and where it breaks — the first conversation is about constraints, not technology.

Start a projectFixed-price checks