The headlines love an adoption number. The OECD’s discussion paper for the G7 is more interested in why the number isn’t bigger.
Its list of barriers isn’t mysterious: skills shortages, financial constraints, weak data and digital readiness, a lack of vendors and use cases built for smaller firms, and — the hard part nobody wants to do — redesigning how the work actually happens.
That last one is the real one. You can buy a seat. You can’t buy a redesigned process off the shelf.
Most AI tooling is still built for companies with a platform team to bend it into shape. A ten-person firm doesn’t have that team. It has one person who already has a full-time job, being asked to also become the change-management department.
Skills and financing gaps get the attention because they’re easy to put a number on. The process gap doesn’t show up in a survey. It shows up six months later, when the tool is still running but nothing about the work has actually changed.
Source: OECD, “AI adoption by small and medium‑sized enterprises,” 2025.
If skills and financing are the visible barriers, what does it take to actually clear the invisible one — redesigning the process itself?