Most small businesses that use AI are spending $100 or less a month. That budget is real, and it is enough for drafts and notes. It is not enough for the work that actually moves a shop in Statesville or Mooresville.
Axios reported exclusive Goldman Sachs data from a survey of 2,474 owners in the 10,000 Small Businesses program. Seventy-three percent of those businesses spend $100 or less per month on AI. In the Washington, D.C. sample, 69% spend $100 or less and 12% pay nothing. In the New York City region, 65% stay at $100 or less.
That is not a rounding error. That is the typical check.
The same survey found 86% of program graduates use AI at work. Goldman’s own write-up of that alumni survey (2,473 responses, fielded August 10–24, 2026) says more than 85% of graduates who responded use AI in some form. Axios later corrected its story to say those businesses are using AI, not fully integrating it. That correction is the whole article.
What $100 a month is good at
Axios is specific about how owners in that sample actually use the tools.
In the D.C. area, the most common use was productivity: taking notes, writing emails, documenting processes. In Texas, 71% of AI users pointed to generating written marketing content more than anything else. In Florida, 77% of users named productivity and 72% named written content.
That is a fair description of a ChatGPT or Claude subscription on a laptop. A Mooresville office manager drafts a follow-up. A Statesville owner turns a job note into a Google post. A Lake Norman shop writes a better about page than they would have on a Sunday night.
Keep doing that. It is cheap, and it is not nothing.
What it will not do: read last week’s jobs, apply your markup, stay inside your service zip codes, or write the quote into the system you already run. A $100 tool does not know your customers unless someone pastes them in. Paste is not a process.
When the cheap stack stops being enough
Goldman’s story is not really about chat windows. It is about owners who wired models into the business.
Marie Hunter, who runs a fuel distribution firm outside Atlanta, started with an agent on email. She and a colleague then used it on a tank-upgrade idea. Instead of costing out a few custom tanks, the agent came back with a plan to upsell a new product line. She calls it a force multiplier. That is not a marketing-copy prompt. That is a tool with a job.
Bob McClure of McClure’s Pickles integrated professional language models into the company’s enterprise resource management program. Food-recall traces that took 30 to 45 minutes of manual data entry now show the path from ingredient to jar in less than two minutes, plus a dashboard for shipments, warehouse stock, and pH. He also built a maintenance monitor, an accounts-payable drafter, and a label checker. Goldman is blunt that writing the instructions takes iterations, and that there are bugs. The point is the integration, not the chatbot.
Forty-two percent of the graduates in that survey use AI for financial analysis and business planning, more than double the 2024 rate. More than half use it to develop or research new projects and products. Asahi Pompey of Goldman put the gap in one line: the challenge is no longer whether small businesses will adopt AI. It is how they harness it fully and turn experimentation into lasting results.
A Troutman HVAC company that pastes invoices into a free tool is in the first group. A Troutman HVAC company whose quoting rules live next to the job records is in the second.
What “spend more” actually means here
It does not mean a bigger ChatGPT bill.
It means connecting the model to the jobs, invoices, and customer records you already have, with a person on anything that moves money or a calendar. That is AI engineering, usually sitting inside Laravel development if you need a real application, not a pile of tabs.
Spend more when:
- The same prompt gets run ten times a week and someone still has to copy the answer into QuickBooks, Jobber, or a spreadsheet.
- Customer names, photos, or prices are going into a personal account because there is no approved place to do the work.
- You want the tool to draft a quote from last year’s similar jobs, not from a blank chat.
- Two people on the team are using two different tools and neither one matches the way you actually price.
Stay at $100 when the job is “write this email” or “turn these notes into a paragraph.” That is a good use of a cheap seat. AI optimization is a different bill entirely: showing up when someone asks ChatGPT who to hire in Iredell County. Do not confuse visibility with operations.
A simple split for the next 90 days
Keep the $100 tools for writing. Pick one operational job that those tools cannot see: after-hours intake, quote drafting from past work, or a recall-style trace through your own records. Put that job in a system you control. Measure minutes saved for two weeks. If the number does not move, the job was too vague. Do not add another subscription.
systemsevendesigns helps owners around Statesville, Mooresville, and Lake Norman draw that line: keep the cheap tools for drafts, and spend the next dollar on an agent that sits in your own data. If you want that line drawn, bring last month’s AI charges and the task that still eats Monday morning. That is enough to start.