Goldman Sachs asked 2,473 owners from its 10,000 Small Businesses program about AI in August. 86% use it. 15% have it fully integrated into their core operations. Most of the rest open it often, too: 63% of the owners who use AI use it every day.
So the gap sits between a chat window the owner opens every morning and a step in the business that runs whether or not anyone remembers it.
What the gap looks like
Goldman's own numbers sketch it. Most owners who use AI (86%) reach it through a chatbot like ChatGPT, Claude or Gemini; 54% use AI features built into software they already have. Only 12% use anything custom-built. And 74% of owners who use AI have no formal policy for how their employees use it.
Goldman calls the next step taking AI from an owner's tool to a business-wide asset. Right now it's used often, by individuals, and rarely written into how the business runs.
Why tried tools don't stick
Goldman's survey names one barrier: among owners who use AI, 39% outside rural areas and 47% in them cite a lack of technical expertise. Before blaming expertise, here are the three reasons I'd look for first.
Nobody owns the task. “We should use AI for follow-ups” isn't a task anyone owns. When it's everyone's, it's the owner's on a good week and nobody's on a busy one.
Nothing is measured. If you didn't count how long estimates took to follow up before, you can't tell whether AI changed it, and a tool that can't show it helped is the first thing dropped in a busy month.
It sits beside the work. Copying notes into a chat window, then pasting the answer back, is an extra step. Extra steps get skipped. A draft that's already waiting in the drafts folder doesn't.
Move one task from tried to running
Pick one task that happens every week. Here's how it goes for an illustrative HVAC office:
- The task: a follow-up email two days after every estimate. Today it happens when someone remembers.
- The owner: the office manager, by name. She decides what a good email looks like and signs off on changes.
- The before number: last month, 41 estimates went out and 17 got a follow-up within two days. Count it from the estimates list in your field service software, or from the sent folder.
- In the work, not beside it: the estimate notes produce a draft in her drafts folder each afternoon. She edits and sends. No chat window.
- The check: a draft can carry over a price from an old estimate and still read perfectly well. She checks every dollar figure and date before sending, and anything about warranty goes to the owner.
- Four weeks later: count again. If the number moved, keep it and pick the next task. If it didn't, drop it.
Check the no-AI answer first. Your field service software may already send estimate follow-ups, and a template does this one without AI. AI earns its place when each email has to mention what the tech saw on the job. If a template does the job, that's a fine result.
To find the task, how AI can help a small business has the usual candidates and what to check in software you already pay for. My roadmap goes through your workflows and decides what to build first, including where the AI step sits inside the work: $1,500 · ~2 weeks · credited in full if you continue. It's a plan; the building comes after. If a template or a feature you already pay for covers a task, the plan will say so.