Two-thirds of Scottish firms use AI — but only lightly

TL;DR:

• 64% of Scottish businesses used AI in the past three months, up from around half the previous quarter, according to the Fraser of Allander Institute’s latest Scottish Business Monitor.
• A majority of firms in every major sector now use AI, peaking near three-quarters in services, but most report only low-to-moderate use.
• More than half of construction and manufacturing firms said they were using AI for the first time.

The survey of over 300 firms points to a familiar pattern: adoption is broad but shallow. Most businesses lean on AI as a supporting tool for day-to-day tasks rather than embedding it in operational processes. Of those using it, around three-quarters apply it to role-specific work such as graphic design, specialist advice and business planning, and seven in ten use it to support decision-making and planning.

Mairi Spowage, director of the institute, framed the shift plainly: “The key question is no longer whether businesses are using AI, but how deeply the technology can be integrated into business processes to support productivity and long-term growth.”

The same story, nationwide

The finding mirrors the UK-wide picture almost exactly. Only yesterday, ONS data showed British firms adopting AI widely but shallowly — the same gap between headline uptake and genuine integration. That two independent datasets, one Scottish and one national, land on the same conclusion in the same week suggests the shallow-adoption plateau is structural, not regional.

It is also the productivity crux. Light-touch use captures little of the efficiency gain that justifies AI investment, and the Monitor found firms remain cautious about longer-term spending despite improving sentiment — held back, Spowage noted, by economic and political uncertainty rather than access to finance.

Looking forward

For Scottish SMEs, the adoption milestone is less meaningful than the depth question behind it. The firms pulling ahead will be those that move AI from occasional assistant to embedded process — and the data suggests most have not yet made that leap. Closing the gap between using AI and running on it is where the productivity dividend, so far largely unrealised, actually sits.