AI adoption is rising. But are businesses actually integrating it?

14 September 2026

AI adoption is rising quickly across UK businesses, but using an AI tool is not the same as integrating AI into the way a business works. The next challenge for SMEs is turning experimentation into useful, repeatable processes.

AI adoption in UK businesses is rising quickly. According to the Office for National Statistics, around 35% of UK businesses with 10 or more employees reported using at least one AI technology in June 2026, compared with around 12% in late 2023. That is a significant increase in less than three years, and it would be easy to look at the headline number and conclude that AI is rapidly becoming embedded across UK businesses.

The detail tells a more complicated story, over the same period, the average number of AI technologies being used by businesses that had adopted AI increased only modestly, from around 1.4 to around 1.6. The ONS describes AI adoption to date as relatively shallow and suggests that this limited increase in the depth of adoption points towards relatively limited transformative impacts for most businesses so far.

There is an important distinction here Using AI is not necessarily the same as integrating AI into a business.

Opening ChatGPT to research something, using Copilot to summarise a document or asking an AI assistant to help draft an email all count as AI use, but they do not necessarily mean the technology has become part of a repeatable business process. For SMEs trying to work out what AI adoption should actually look like, that distinction matters.

What does AI adoption actually mean?

One of the difficulties with measuring AI adoption is that the term covers an enormous range of activity. A business might have employees occasionally using a general purpose AI assistant, while another might have AI built into its customer service platform, finance system, CRM or internal workflow, both businesses could reasonably say they use AI but the level of integration is completely different.

The UK Business Data Survey 2026 gives us a useful indication of how much AI has actually made its way into existing business systems. Among businesses using AI, only 21% said their AI tools were integrated with existing systems. There was also a clear difference by business size, among AI using businesses, 57% of large businesses reported having AI integrated with existing systems, compared with 31% of medium businesses, 31% of small businesses, 27% of micro businesses and 18% of sole traders.

The survey defines integration broadly and includes examples such as Microsoft Copilot within Microsoft 365, AI embedded in CRM or finance systems, and AI features within workflow or productivity platforms. Even with that relatively broad definition, most businesses using AI are not yet reporting integration with their existing systems, this suggests much of today's AI adoption is still happening at the level of individual tools and individual tasks.

Experimentation is still useful

There is nothing inherently wrong with trying thinks out, in fact, experimentation is probably a sensible part of working out where AI genuinely helps. Businesses need to learn what the technology is good at, where it produces unreliable results, which tasks people actually want help with and whether the time saved is meaningful enough to justify changing a process. The mistake is assuming that experimentation and adoption are the same thing.

An employee finding a useful prompt for summarising documents is an experiment, but a team agreeing when AI should be used to summarise documents, deciding which documents are appropriate, selecting an approved tool, establishing how outputs are checked and making that process part of normal work is closer to adoption. The technology might be identical in both examples, but what has changed is the process around it.

Integration is not just a technical problem

When people hear the word integration, it is easy to think immediately about APIs, databases and connecting software together. Whilst those things can matter, genuine AI integration is broader than technical connectivity. For AI to become part of a business process, the organisation needs to understand what the process is trying to achieve, where AI should contribute, what information it needs, what happens when the output is wrong and who remains accountable for the result. That becomes particularly important when AI moves from helping somebody complete a task to influencing a workflow used by multiple people.

Imagine a business using AI to help process incoming customer enquiries, at the simplest level, an employee might paste an enquiry into an AI assistant and ask for help drafting a response. A more integrated version might classify incoming enquiries automatically, identify the relevant customer record, retrieve useful information, prepare a suggested response and route it to the appropriate person for review.

The second version could potentially save more time, but it also introduces more questions:

Where does the customer data come from?
What information can the AI access?
How reliable is the classification?
What happens when it gets something wrong?
Which actions can happen automatically?
Where is human approval required?
Who owns the process? Those are business process questions as much as technology questions.

Why is scaling AI difficult?

The Department for Science, Innovation and Technology's AI Adoption Research provides some useful insight into the barriers businesses are encountering. Among businesses already using AI, 54% said limited AI skills, expertise or knowledge were hindering or had previously hindered wider adoption across the organisation. Another 30% said they had not identified a use for AI, while 26% said AI projects were too complex or difficult to integrate and scale. Data also appears as part of the problem, with 23% reporting that data complexity was hindering wider AI adoption.

These numbers are useful because they show that the challenge is not simply getting access to AI technology. Businesses can already access extremely capable AI tools relatively cheaply and, in many cases, through software they already use, the harder work is understanding where those tools fit into the organisation. That requires knowledge of the process, access to the right information, people who understand enough about the technology to use it appropriately and a clear reason for changing the way work is currently done.

Data becomes more important as AI becomes more integrated

The relationship between AI adoption and data is particularly important. A general purpose AI assistant can be useful without having access to much company information, it can help someone structure ideas, rewrite text, explain a concept or summarise information provided by the user.

More integrated AI usually needs more context. Just like a new team member joining the business, if AI is expected to answer questions about customers, analyse operational performance, support purchasing decisions or help employees find internal information, it needs access to relevant business information.

That point often quickly exposes existing data problems. Customer records might be incomplete, product information might exist across several systems, different departments might define the same metric differently, and important information might live in spreadsheets owned by individuals rather than in systems the wider business can reliably access.

AI does not make those problems disappear, in most cases, it makes them more visible.

The UK Business Data Survey found an interesting association here, businesses with AI integrated into existing systems were more likely to say they analysed data and collected data than businesses whose AI tools were not integrated. That doesn't prove that better data practices cause better AI integration, or that AI integration causes better data practices, but it does reinforce how closely the two issues are connected. If a business wants AI to become part of its operations rather than remain a standalone assistant, the quality and accessibility of its data increasingly matters.

Not every AI experiment needs to become a system

There is another important point that can easily get lost when discussing AI adoption, not every useful AI task needs to become a fully integrated process, sometimes using an AI assistant occasionally is perfectly adequate. If somebody uses AI twice a month to help structure a presentation, there may be very little value in building a sophisticated workflow around it.

The decision to integrate AI should depend on the business problem:

How frequently does the task happen?
How much time does it consume?
How consistent is the process?
What would better performance be worth?
What information does the AI need?
How much risk is involved if the output is wrong?
How difficult would the process be to integrate? Those questions help distinguish an interesting AI use case from one that is actually worth developing. For an SME with limited time and technology resources, that prioritisation is important because there will always be more potential AI ideas than sensible projects to pursue.

Moving from AI use to AI adoption

The next phase of AI adoption for many businesses is probably not about discovering more AI tools, there's already plenty of them. The challenge is deciding which uses deserve to move beyond individual experimentation and become part of the way the organisation actually works.

That means looking at the task, the process around it, the people involved, the information required, the risks, the expected benefit and what needs to change for the technology to be used consistently. It also means being willing to decide that some AI experiments should remain experiments.

A useful AI strategy should not produce a long list of possible applications and assume they should all be implemented. It should help the business identify which opportunities are valuable enough, realistic enough and safe enough to pursue.

Where should you start?

If people across your business are already using AI, you probably don't need to start by finding more tools ... start by understanding which uses are actually working. Look for tasks where AI is already saving meaningful time, helping people work through information more effectively or improving an existing process. Then look at what would be required to make that use repeatable across the business.

Does it need access to better data?
Does it need integration with another system?
Does the process itself need simplifying first?
Who should own it?
What human oversight is required?
How will you know whether it is actually improving anything?

Those questions move the conversation from AI experimentation towards AI adoption. Node9 helps SMEs identify where AI can genuinely add value, assess the processes and data around those opportunities, and build practical adoption roadmaps based on what the business can realistically implement.

If your business is already experimenting with AI but you are not sure what deserves to move beyond experimentation, start with a conversation.