BCC finds half of SMEs use AI, but job cuts cluster among bespoke users

More than half of the businesses that answered the British Chambers of Commerce’s (BCC) January survey now use AI in some form. The figure, 54%, published by the BCC on 18 March in a report produced with Atos and the University of Essex, compares with 35% in the BCC’s 2025 survey, 25% in 2024 and 23% in 2023.

The jobs story is quieter. Among firms using only generic AI tools, 95% say it has had no effect on the size of their workforce over the past year, and the BCC says 86% of firms report job roles unchanged. “Our research shows AI has rapidly moved from the margins of business to the mainstream,” said Patrick Milnes, head of policy: people and work at the BCC.

The question for leaders buying AI is which part of the research to plan around. The adoption headline is the number to treat with most care. The more useful finding is that changes to headcount and job design cluster in a small group of firms that have moved beyond off-the-shelf tools.

Two counts, one direction

The BCC figure sits well above the official estimate. The Office for National Statistics’ (ONS) Business Insights and Conditions Survey found that 25% of businesses were using some form of AI technology in late December 2025, up 15 percentage points since the question was first asked in September 2023. Among businesses with 250 or more employees the share was 44%, and a further 15% of all businesses said they planned to adopt AI within three months.

Much of the gap is likely to be one of method and sample. The BCC sample of 668 firms comes from its membership network, and the BCC says around 94% of respondents are SMEs. The working paper behind the report, written by the BCC’s head of research David Bharier with Essex economists Ben Etheridge and Paulo Morais, reports unweighted results. It says they should be read as describing “patterns within the responding sample rather than as population-representative estimates for UK SMEs”. Firms that belong to a chamber of commerce and answer a survey about AI may be likelier than average to use it. The ONS, for its part, labels its figures official statistics in development and advises caution.

Both surveys point the same way, though. Adoption is rising quickly, larger firms are ahead and the effect on jobs so far is small. In the ONS data, 4% of businesses using AI said their headcount had fallen as a result. In the BCC survey, the equivalent figure among firms using only generic tools was 3%.

The spread is uneven. In the BCC sample, adoption runs from 78% in IT and digital firms and 70% in creative and media to 45% in manufacturing and 31% in other services. The Essex paper also finds lower use in the North East, Yorkshire and the devolved nations than in southern England.

Where the headcount effect sits

The report’s sharper contribution is a split between two kinds of user. Most firms use generic tools such as ChatGPT or Copilot to draft content, summarize information, analyze data and do research. Around one in 10 have adopted bespoke AI systems built for their organization, which the report says are more likely to sit in core functions such as workflow automation, operational decisions, forecasting or customer service.

That is where the staffing changes are. Of the bespoke users who answered, 21% said their workforce had shrunk as a direct result of AI in the past 12 months, against 3% of firms using generic tools only. The bespoke group is small – 61 firms – but the Essex authors call the difference highly statistically significant and find bespoke adopters roughly three times more likely to have restructured or redefined job roles. They also caution that the results are cross-sectional and that a strong causal reading should be treated with care.

The paper’s own summary puts it plainly: “Headline AI adoption figures may overstate the extent of workforce disruption, since most use remains shallow.” Its authors expect restructuring and shifts in skills to follow as bespoke systems spread.

The training paradox

The finding most likely to surprise employers concerns training. Firms planning to invest in AI-related training were more likely to expect headcount to fall, not less: 14% of those with training plans expected a decrease over the next 12 months, against 4% of firms without them. The authors call this ‘replace-and-train’ and suggest that, for some firms, training accompanies workforce reduction rather than preventing it.

Investment in training is thin in any case. The report cites earlier BCC research showing that only 2% of employers plan significant investment in AI-specific training, while almost a third say they now put more weight on AI literacy when assessing the skills they need. The BCC’s own AI Academy, run with Pair since October 2025, has trained more than 350 individual users, the report says.

Optimism tracks experience. Firms already using AI gave a net productivity expectation of +71 points for the coming year, against +46 for firms planning to adopt and a slightly negative score for those with no plans. That is expectation, not measured return. The Essex paper notes other research finding that most firms report no measurable productivity gains from AI so far.

The BCC’s recommendations are aimed at government: a network of AI champions to help SMEs adopt the technology, an observatory to track AI’s effect on the labor market, Growth and Skills Levy funds for AI literacy training, and tax credits or grants for investment in AI and workforce training. Shevaun Haviland, director general of the BCC, writes in the report’s foreword that “AI is not without its problems”. AI minister Kanishka Narayan pointed to the government’s new Future of Work Unit and its aim to help 10 million people build practical AI skills. “The next step is making sure companies of every size feel confident using it day‑to‑day,” he said.

What buyers should take from it

For SME leaders, the practical lesson is to treat these as two separate purchases. Generic assistants are now a mainstream productivity tool, and both surveys find AI has rarely changed headcount so far. A bespoke system is a different decision: it is where job roles are being redesigned, and it should be bought with a plan for the people and skills around it, not only the software. Either way, firms should measure productivity against a baseline before scaling, rather than relying on the confidence of early users. Buyers who read the 54% as proof that they are falling behind are reading the wrong number.