Between May and June, Anthropic ran free half-day workshops for small businesses in ten US cities: Chicago, Tulsa, Dallas, Hamilton Township NJ, Baton Rouge, Birmingham, Salt Lake City, Baltimore, San Jose and Indianapolis. More than 1,000 owners showed up and 635 filled in an exit survey. The write-up went online yesterday.
Who attended is already telling. 80 percent of registrants run companies with 5 to 50 employees. The industries: construction, manufacturing, logistics, trades, professional services. The part of the economy that touches physical things - not the part AI articles usually get written about.
Four findings
The top use case was not writing. It was operational reporting and automating repetitive tasks. The biggest worry was data security and governance. The biggest barrier was the learning curve for new tools. And what helped attendees most was learning from each other.
That last one is the finding I keep turning over. Not the docs, not the tutorial - the owner two tables over who had already solved the same problem. Anthropic is acting on it with an SMB trainer programme, and part two of the tour starts in Boston on 16 September.
Two examples that stick
A firm that handles fuel tax filings for trucking companies cut its error rate from 7 percent to zero. An industrial systems integrator now identifies parts in seconds instead of hours.
Neither is a spectacular story. Both are jobs that used to eat half a day and that nobody enjoys. That is where automation actually starts to bite, and it rarely gets written up, because it does not sell as a breakthrough.
One quote from the post stays with me: “You were able to help me do the things I was scared of doing with AI.” That is the CEO of a 30-person manufacturer. Scared, not uninformed. The difference decides what kind of help works.
Why a model vendor runs workshops in Tulsa
The blunt reading: this is sales. Anthropic sells to companies, small business is a huge under-served market, and whoever runs the training picks the tool.
The other reading does not contradict it. The bottleneck for AI in small business has been the same for two years, and it is not model quality. It is that nobody is there to show you how to start. Anthropic measured that instead of asserting it - 635 surveys is not a large study, but it is more evidence than most vendors bring to this question.
It does not transfer one to one outside the US. Different privacy regimes, different company sizes, different support structures. The peer-learning finding probably holds anywhere, though. If you sit in a trade association or a business network: that is your opening, and you do not need a vendor from California to use it.
Sources: Anthropic: What 1,000 small business owners taught us about AI