SMBs and AI: 49% got a bigger bill than expected, and leaders now review more than they produce
A Pax8 survey of 1,150 SMB leaders (US, UK, Germany, Sweden) points to a gap between AI use and AI oversight. The panel is Anglo-Saxon and Nordic; France is not in it.
According to SMBtech (October 13, 2026), reporting on the Q3 Pax8 SMB AI Pulse Report, 35% of AI-engaged businesses found tools or agents running without official approval in the past six months. In the US, only 36% of those that built their own solutions have a documented review. Among businesses that paid for AI, 49% received at least one bill or usage charge higher than expected.
The time saved is real (69% say it freed up capacity), but it moves: 55% of decision-makers spend at least three hours a week reviewing, editing or fact-checking AI-generated work. And 70% would pay at least 6% more for better control over how their data is stored and used. For an agency or a network, the useful question is not "which AI?" but "who signs it off, and what does it cost?".
For AppH
- An assistant bounded by the agency's own rules cuts down what needs reviewing: it prepares replies and reminders, and the agency sees what was proposed before anything is sent.
- A human stays in control: the agency approves every rule and every text, and the cost is known in advance, with no per-resolution meter.
Against / the honest limit
- Reviewing is still work: a well-bounded assistant reduces the volume to check, it does not remove it.
- The survey covers businesses with 5 to 499 employees in four countries, not French domiciliation networks; it comes from a cloud marketplace provider, and we have no measurement specific to our trade.
The AI that saves time is the one where you know who approves it and what it costs. Without that, the work just moves to proofreading.
Reviewed by a human at AppH