AI use has gone wide but not deep, and the missing piece is people who can wire it into the work

The UK government published the second wave of its AI adoption survey on 29 October, based on 3,503 interviews with private-sector businesses that employ five or more people. The headline is a doubling: 32% were using at least one AI technology, including pilots, against 16% a year earlier. Use rises with size, from 27% of micro firms to 52% of large ones, and information and communication firms lead at 61%.

The questions put to users change the picture. Only 14% of firms using AI rate the tools of high importance to their operations; 44% give a low rating. Asked about operating costs, 64% of users report no impact, 24% a reduction and 9% an increase. Half say employee productivity has risen since they adopted AI, though the report describes those gains as generally modest. Among current users that could estimate their spending, 43% spent nothing on AI in 2025.

FDE AI reads that last figure as the key to the rest. A business that only uses a free assistant for drafting emails can count as an adopter without having changed a process, a role or a system. Put the two headline numbers together and fewer than one in twenty UK businesses with five or more staff both uses AI and rates it highly important, our own calculation from the survey's figures.

The Office for National Statistics, using a different survey of firms with ten or more staff, finds the same split. Self-reported AI use has risen from around 12% to around 35% since late 2023, but only 10% of those users describe their use as extensive.

A US Census Bureau working paper, drawing on survey data for November 2025 to January 2026, found that 18% of firms used AI in a business function, or 32% when weighted by employment. Among those adopters, 57% use it in three or fewer business functions. AI-related reductions in employment were reported by only 2% of firms. The authors also report a robust positive link between a firm's commercial performance and how broadly it has integrated AI, an association rather than proof of cause.

Statistics Canada reports that 19.2% of businesses used AI to produce goods or deliver services in the second quarter of 2026, up from 12.2% a year before. Its size breakdown shows how differently large and small firms build capability. Among firms with 100 or more employees, 68.1% trained existing staff, against 24.0% of firms with one to four employees. Large firms also used external consultants or vendors far more often, 30.2% against 10.7%, and hired people with AI skills four times as often, 32.8% against 8.2%.

Europe and Australia show rising use, measured on different definitions. Eurostat counts 20% of EU enterprises with ten or more employees using AI in 2025, up from 13.5% in 2024. Australia's National AI Centre put AI adoption among small and medium-sized businesses at 43% for December 2025 to February 2026, and named trust as the main reason for holding back. This piece draws on UK, North American, EU and Australian data; readers in India, Singapore or the Gulf should treat the pattern as a test to run on their own firms.

These surveys define adoption differently, so their percentages should not be lined up as a league table. What they share is the direction: wide use, narrow integration, and a large-firm lead in training and outside expertise.

The UK survey asks what holds firms back. Limited AI skills, expertise or knowledge is the most common barrier, cited by 67% of firms using AI and 78% of firms planning to. Among non-adopters, 65% cite the lack of a clear business case. And 81% of firms using AI keep some level of human input or checking on its outputs.

Taken together, these place the barrier in the work around the model. Someone has to decide which process to change, connect the tool to the data and systems that process depends on, define who checks the output, and show the result in numbers a finance director accepts. Firms can buy access in an afternoon; a redesigned claims process or month-end close takes longer.

FDE AI reads the Canadian size gap as one sign of how firms close that gap. Large firms train their own people and bring in consultants, vendors and hired specialists far more often than the smallest. The forward deployed engineer (FDE) model, an engineer embedded with a client to take a system from pilot to production, is one way of buying that capability. It is not the only one, but FDE AI's reading is that capability, however it is sourced, is what separates firms that use AI from firms whose operations rely on it.