Fact: In a 29 September 2026 article, OpenAI said that roughly 4 million employees at businesses with fewer than 500 people worldwide used its tools during one week, from 9 to 15 September. In the report it cites, nearly one in five of those users worked at a business with fewer than 10 people. OpenAI also said that in August, work in finance and accounting, supply chain, legal and compliance, and marketing represented at least three times the share of small-business interactions with AI agents that those areas represented for larger businesses.
These figures are data published by OpenAI, not an estimate for every small business. They nevertheless point to an important management issue: small firms often turn to AI for work where they lack in-house specialists and where the owner or a small number of key employees feel the pressure most.
Do not begin with “where could we use AI?”
That question is so broad that it can produce a long list of ideas while no one job gets completed properly. A better starting point is one recurring job that slows the team down, has clear inputs, and can be checked within a few weeks.
For example, a small trading business may spend many hours each week answering the same product questions, comparing competitor prices, or preparing content for sales campaigns. Instead of immediately buying a large system, it could select one job: create a weekly price comparison for 20 priority products. The inputs are a product list and permitted public sources; the output is a table with an update date, data sources, and items the responsible person must verify.
Choose the job with three criteria
First, make the output visible. The result should be a draft email, a screened lead list, a comparison table, or a meeting summary, not a vague goal such as “improve productivity.”
Second, name the person responsible for checking it. AI can create a first draft, but someone needs responsibility for accuracy, tone, sensitive data, and the decision to use the result. For customer-facing content, the reviewer must understand the brand. For financial figures, the reviewer must know the source data and accounting rules.
Third, use a measure small enough to follow. It could be preparation hours saved, the share of drafts used after revision, customer response time, or errors found before sending. The measure does not have to be perfect. It does need to help the business decide whether to continue, change the method, or stop the test.
Turn the test into a small process
An AI test becomes useful when the team records how to repeat it. For each job, keep a short page with the objective; data or documents that may be entered; the prompt used; review criteria; the approver; and where the result is saved. This is not heavy process. It prevents a situation where one employee can do the work but nobody else knows where to begin.
For instance, if a marketing team uses AI to draft customer-care emails, the process should specify which information may not be entered into the tool, who checks promotional promises, and where the final version is stored. When a result falls short, the team can tell whether to improve the input data, prompt, or review step instead of concluding vaguely that AI “does not work.”
Management interpretation
Small businesses do not need a large collection of AI tools. They need a few dependable workflows. The figures published by OpenAI suggest that demand often appears in work supporting specialist functions. Lasting value arrives only when a company connects the tool to a specific job, a responsible person, and a clear measure.
That is why the most useful next question is not “what can AI do for everything?” It is: “Which recurring, time-consuming job is safe enough to test for two weeks?” The answer usually creates a much more practical starting point.
Source: OpenAI, Helping small businesses put AI to work, 29 September 2026.
