Recent AI developments that matter for small and mid-sized businesses.
Short, sourced notes on how SMBs are putting AI to work, across operations, marketing, finance, and more. Updated weekly.
Policy/Riskcross-industry
A third of SMBs using AI found tools running that nobody approved
Pax8, which sells AI and cloud software to SMBs through IT service providers, surveyed 1,150 technology decision-makers at companies with 5 to 499 employees in the US, UK, Germany and Sweden. It found that 35% of AI-using businesses discovered tools, workflows or agents running without official approval in the past six months, rising to 42% in the US. Pax8 also found that 67% of AI users have built at least one solution themselves, while only 36% of US builders put that work through a formal documented review. These are Pax8's own figures and have not been independently verified, and the report's conclusion that IT partners should help govern SMB AI happens to describe what Pax8 sells.
The pattern still matches what we see. Shadow AI usually starts with a capable employee solving a real problem, and the risk is that nobody else knows the workflow exists when it breaks, or when a client asks what happened to their data. The practical move is an inventory before a policy. Ask each team what they have built, write down what data it touches, then decide which ones need a review step.
A trucking owner ran her own fleet data through AI and found $53,000 in idle time
Pam Polyak, third-generation owner of Polyak Trucking in Wisconsin, fed her Samsara fleet data into Claude and asked it to flag any driver logging more than 30 minutes on duty without driving at the terminal. One driver was averaging more than 110 minutes per shift. Polyak told FreightWaves that the pattern added up to roughly $53,000 in hourly labor, and that the analysis itself took about three minutes. The same approach surfaced a $50-per-load toll gap between two drivers running the same lane in opposite directions. Both figures come from Polyak's own account in an interview and have not been independently audited.
The tool is the least interesting part of this story. The data already existed, sitting in a telematics system the company was paying for, and nobody had the hours to read it. That is the most common AI win at this size. Start by listing the reports your systems already produce that no one opens, then pick the one where a missed pattern would cost real money. Keep a person on the judgment call: the system flags the behavior, you decide what it means.
