83% of large enterprises have deployed AI. 18% of businesses under 50. Why is it going top-down?
We (SigmaPointPi) spent three years meeting people before building anything. 500+ founders, operators, accountants, tax preparers and lawmakers, face to face.
I went in expecting arguments about price. That is not what I got...
They are really hesitant. Specifically of being wrong in front of a customer, an auditor, or the state. 41% cite cost, 37% say they do not know enough, 29% do not trust accuracy, and 38% of the ones who have adopted nothing name data privacy outright.
Then it stopped being about small operators. I sat with an MIT nuclear engineering professor and asked where AI is permitted near high-consequence physical systems. He mentioned that one of his own students had just founded an AI nuclear startup, and he laughed. He knows exactly how far a working demo sits from something you would place near a containment loop.
Experts hesitate hardest inside their own field, because they can price the failure precisely. Which explains a lot about why adoption skews young.
83% deployment above 5,000 employees, 18% below 50. Micro-businesses at 23% against 67% for larger SMBs. A 44-point spread.!!!
Every general purpose technology that actually rearranged the economy went the other direction. PCs, the internet, mobile. Cheap, bottom-up, reaching the people with the least slack first.
This one is arriving top-down. (TOP-DOWN)
is that a distribution problem somebody can solve, or is it just what happens when the technology is capital-intensive at the frontier?
Replies
I don't think it's purely a capital problem, it's more that big companies already have a safety net for being wrong - compliance, legal, a QA layer - so an AI mistake gets caught before it reaches the customer or the auditor. a solo accountant or small law office doesn't have that layer, so an AI error becomes their personal error with their name on it. PCs and mobile never asked you to trust an output, they just gave you a tool to produce your own. AI is the first general purpose tech that asks the small operator to vouch for something they didn't fully create, and that's a much scarier ask when you have no one to catch it if it's wrong
I think part of the gap is that adopting AI isn't just “pay for a tool and start using it.”
Larger companies usually already have structured data, documented workflows, APIs, approval processes, and people who can integrate AI into those systems.
A small business may still be running important operations through WhatsApp, spreadsheets, email, and knowledge that exists only in the owner’s head.
So even when the AI itself is cheap, making the business ready to use it safely can be expensive.
That makes me wonder if the bigger opportunity isn’t cheaper AI, but products that meet small businesses where their workflows already are instead of asking them to become “AI-ready” first.