DSIT AI Adoption Research: What It Means for Business

The Department for Science, Innovation and Technology has published AI Adoption Research, a detailed assessment of how UK businesses are using artificial intelligence based on 2025 fieldwork, using 3,500 survey interviews and 100 follow-up qualitative interviews.

The report matters because it gives business leaders a grounded baseline. Adoption is not yet universal, most businesses have no active plan to use AI, and the most common business uses are still practical language tasks rather than highly autonomous systems. That is useful context for teams deciding where to invest, where to slow down, and what evidence they should expect before scaling.

This is not a model release or a product announcement. It is a research paper, so the business value is in using its findings to test adoption assumptions, prioritise training, and plan governance.

What the research covers

DSIT commissioned IFF Research and Technopolis Group to assess business use of AI, the barriers that affect adoption, and the early impact reported by businesses. The fieldwork took place from 12 February to 2 May 2025 and covered private sector businesses with at least five employees.

A practical adoption survey

The survey asked businesses whether they use AI, where they use it, what stops them adopting it, and what impact they have seen. It also explored categories such as natural language processing, text generation, computer vision, machine learning, and agentic AI. The result is a useful view of operational adoption rather than a forecast based only on market sentiment.

A limit to remember

DSIT notes that the survey approach will not fully capture shadow AI adoption. In practice, some staff may already be using public tools or embedded software features outside formal approval routes. That matters for governance because a business can have real AI exposure before it has an official AI programme.

What adoption looks like

The headline finding is that AI adoption remains modest. Around 1 in 6 businesses, or 16%, are currently using at least one AI technology. A further 5% say they plan to adopt AI in future, while 80% are neither using AI nor planning to use it.

Use is uneven by size and sector

Larger businesses and mid-sized businesses are more likely to be adopters than micro firms. Adoption is also higher in information and communication, finance and real estate, and business services and administration. This pattern is unsurprising: these sectors often have data-heavy workflows, clearer software budgets, and more obvious generative AI use cases.

Most adopters are using language tools

Among businesses already using AI, 85% are using natural language processing and text generation. That points to familiar work such as drafting, summarising, transcription, analysis support, and code generation. By contrast, agentic workflows are still much less common, which fits with the extra governance, data access, testing, and operational design they require.

Barriers holding businesses back

The report separates common barriers from the barriers businesses rate as most significant. That distinction matters. Many organisations say they have not identified a need for AI, but where barriers are felt strongly, ethical concerns, cost, and unclear regulation carry more weight.

Need and skills come first

Across all businesses, the most common barrier is a lack of identified need, cited by 71%. Limited skills, expertise, or knowledge follows at 60%. For teams planning adoption, this suggests the first step should usually be workflow discovery and staff capability building, not buying another tool.

Governance concerns remain material

Among businesses that had cited those barriers, ethical concerns were rated significant by 80%, high costs by 76%, and unclear regulation by 72%. Those numbers support a practical point: AI programmes need clear controls for risk, value, and accountability. A well-written prompt library is useful, but it does not replace governance, evidence, and ownership.

Productivity gains need context

Businesses already using AI are generally reporting operational benefits. DSIT says 75% reported improved workforce productivity and 57% reported new or improved processes or operations. Over half also reported an increase in employees' overall productivity since adopting AI.

The revenue story is more cautious. More than three quarters of businesses using AI reported no revenue change yet, while 12% reported an increase. That does not mean AI has no commercial value. It means early value is more visible in time, capacity, quality, and process measures than in immediate revenue uplift.

For business leaders, the sensible lesson is to set measurable but realistic success criteria. Track avoided rework, cycle time, output quality, and staff adoption before assuming AI will directly move revenue. Also treat the figures as self-reported estimates, not independently verified productivity measurements.

Trust, safety, and human oversight

The research identifies data security and output accuracy as common safety concerns. That aligns with what we see in client conversations: businesses are often interested in AI, but they need confidence that outputs are checked, sensitive data is protected, and mistakes are caught before they affect customers or operations.

Human checking is already common

Among businesses using AI, most reported at least some input or checking of AI outputs or decisions, and only a very small minority reported no checking. That supports a human-in-the-loop approach for early adoption, especially where outputs affect customers, finance, compliance, or safety.

Accuracy is a governance issue

Output checking should not be left to individual preference. Teams need review thresholds, data rules, escalation points, and acceptance tests. That is particularly important where a model may produce a plausible but wrong answer, sometimes described as a hallucination.

How businesses can use the research

The DSIT report should be read as a planning baseline, not a recommendation to adopt a particular platform or vendor. Its value is in showing where adoption is currently concentrated, where businesses are getting stuck, and which controls are becoming important for responsible AI deployment.

The most relevant lesson for business leaders is that adoption should be tied to a specific business need. Start with high-friction workflows, define how success will be measured, put clear review rules in place, and build staff confidence before moving into more autonomous systems. That is a more reliable path than treating AI adoption as a generic technology project.

If you would like to discuss AI adoption planning or governance, contact us at enquiries@coaleypeak.co.uk.