Summary

• AI Scotland has received £1.8 million for a second year of SME support.
• The previous £1 million programme supported more than 1,000 businesses, according to government figures.
• The programme combines mentoring, workshops, grants, implementation planning, and technology testing.

Governments trying to raise productivity through artificial intelligence face a harder task than persuading companies to experiment with chatbots, because smaller businesses often lack the technical staff, data infrastructure, procurement capacity, and implementation experience needed to turn a promising trial into a working system.

Scotland is extending one attempt to bridge that gap, with AI Scotland receiving £1.8 million of government funding for a second year of support aimed at small and medium-sized businesses. The programme combines mentoring, workshops, grants, implementation planning, and opportunities to develop and test AI systems around individual business requirements.

The funding follows £1 million allocated to the previous National AI Adoption Programme for SMEs, which the Scottish Government says supported more than 1,000 businesses. Delivery involves Scotland’s enterprise agencies, Business Gateway, and The Data Lab, combining business support with technical expertise rather than treating adoption purely as a technology-purchasing exercise.

The more useful test will be whether participating companies identify processes where AI produces sufficiently reliable operational improvements to justify the cost and disruption of deployment. Smaller organisations have less room than large companies to absorb experiments that consume management time without producing a commercial return.

Adoption begins with choosing the problem

Many businesses already have employees informally experimenting with generative AI, but organisational use requires a more deliberate sequence of decisions. Companies have to determine which processes are suitable, what information an application needs, how outputs can be checked, and whether the system connects cleanly to existing software.

AI Scotland is designed partly around that earlier stage, helping businesses identify opportunities and plan implementation before development and testing. The structure acknowledges a recurring problem in enterprise AI projects: enthusiasm for the technology can arrive before an organisation has defined the operational constraint it expects to remove.

The Scottish Government points to Border Biscuits in Lanark as one participant, saying the manufacturer has applied AI to data analysis, production-line monitoring, and sales forecasting. Those are conventional business functions compared with autonomous agents or bespoke foundation models, but they are also bounded problems where a smaller business can judge whether the system produces a measurable improvement.

Manufacturing demonstrates why implementation support matters. Forecasting depends on usable historical information, while production analysis requires reliable data from factory systems and a clear understanding of how recommendations fit into existing processes. Access to an AI model cannot compensate indefinitely for weaknesses in the underlying data.

SMEs face different adoption economics

Large organisations can create specialist AI teams, fund several pilots simultaneously, and negotiate directly with major technology suppliers. Smaller companies are more likely to depend on external advisers and features built into software they already use.

Public support can therefore influence adoption without requiring government to develop the underlying models. Assistance with vendor evaluation, data preparation, governance, process redesign, and implementation can reduce the risk that companies spend scarce resources on systems that do not solve a defined business problem.

The programme’s success cannot, however, be established through workshop attendance or the number of companies receiving support. A stronger productivity measure would examine whether participants continue using their systems after the intervention ends, whether processes become cheaper or faster, and whether businesses gain capabilities they can apply to later projects.

The first year’s reach across more than 1,000 businesses indicates substantial demand, but the latest announcement does not provide detailed outcome data demonstrating the resulting productivity gains. The second year therefore offers an opportunity for evaluation to move beyond participation figures.

AI policy moves into ordinary businesses

The programme also reflects a wider change in technology policy. As general-purpose models become commercially available, economic impact depends increasingly on whether AI diffuses into organisations that are not themselves technology companies and are unlikely ever to train foundation models.

That diffusion is uneven because adoption depends on management capability, cybersecurity, data quality, software architecture, employee confidence, and the suitability of existing processes. Informal use can also spread before companies establish approved tools or policies.

A programme built around mentoring and implementation addresses part of that gap, although it also risks becoming another broad innovation initiative whose results become difficult to measure once grants and workshops have ended.

The £1.8 million commitment is modest beside national spending on computing infrastructure and AI research, but it tests a different policy problem: whether targeted intervention can help ordinary companies move through the less visible work of selecting a process, preparing information, deploying a system, training employees, and deciding whether the resulting economics justify keeping it.