What you are holding at the end of this page is a four-document set you can put in front of a board, an auditor or a supplier without further editing. Page one is a single paragraph that defines what your organisation treats as AI use, signed and dated by a named executive. Page two is an inventory of every AI system already running, with the decision each one changes and the person who has that decision. Page three is a sourcing decision for the next system you buy, with three delivery routes set against the obligations each one leaves you holding. Page four is the retirement trigger for each system, written before it becomes active. The rest of this page is the work that creates those four pages, arranged as the questions a person actually asks, in the order they arrive.
What does artificial intelligence in business actually mean?
AI in business is the use of a defined set of technologies to change a decision or an action, and the technology list is the least interesting part of that sentence. IBM's definition of artificial intelligence in business lists machine learning, natural language processing, generative AI, predictive analytics and computer vision: these tools automate work, optimise operations, improve decision-making and drive business value. Four of those five items already sit inside your organisation under a different name. Machine learning sits inside the forecasting spreadsheet the planning team has run for years. Natural language processing sits inside the service desk that classifies tickets before an agent opens them. Computer vision sits inside the camera above the loading bay door. The useful question asks which decisions now run on a model, and whether anybody wrote that down.
The same IBM page has two figures that set up every other part of this page. IBM's research there reports that 79 per cent of executives say AI has improved productivity and will contribute significantly to revenue by 2030, while 24 per cent can clearly see where that revenue will come from. The distance between those two numbers is why this page treats artificial intelligence in business as a management discipline, with the technology purchase as the smaller half of the work.
How many UK businesses actually use AI, and what does the number hide?
The UK figure that remains after a board meeting is the one collected on a consistent basis, and the Office for National Statistics has collected it since late 2023.
Source: Office for National Statistics, Artificial intelligence in UK businesses: 2023 to 2026
| Measure | Latest figure | Observation window |
|---|---|---|
| Businesses with ten or more employees reporting use of at least one AI technology | 35% | June 2026 |
| Same measure at the first wave | 12% | September to October 2023 |
| Average number of AI technologies per adopting business | 1.6, up from 1.4 | June 2026 |
| Information and communication sector reporting AI use | 58% | June 2026 |
| Construction sector reporting AI use | 13% | June 2026 |
| Businesses describing their AI use as extensive | 10% | June 2026 |
| Employees reporting AI use for work or education | 55% | May to June 2026 |
Three things in that table matter more than the headline. The share of businesses using AI has roughly tripled while the average number of technologies per adopter has barely moved, so adoption is broad and shallow, and the ONS warns that on this proxy the transformative impact for most adopting firms remains limited. The sector spread is wide enough to make any national average misleading on its own, because knowledge-intensive industries sit far above construction. Employee-level use runs well ahead of business-level use, which the ONS attributes to informal and ad hoc activity that firm surveys were never designed to capture. If your acceptable use policy was written against the business number when employee-level use runs higher, your first serious AI incident will arrive through a personal account on a work laptop and you will hear about it from a client.
The ONS is explicit that survey design changes the answer. Its article notes that firm-level surveys of senior executives produce much higher adoption rates than bottom-up worker surveys, that larger firms report higher rates than smaller ones, and that measurement of AI use remains an evolving area in the UK and internationally. When a vendor quotes you an adoption rate, the first question is which survey produced it and what that survey counted.
Where does AI in business actually show up in the work?
Improving existing operations is the most common purpose for AI use across every employment size band, according to the ONS. Providing or personalising products and services is the second most reported purpose, and developing new products or exploring new markets comes after that. That ordering should reset expectations for any AI programme sold internally as a growth play.
The functions where this work lands are ordinary. IBM lists marketing, sales, customer support, finance, procurement, information technology, software development, human resources and supply chain operations as the places AI is already operating. Case studies are most useful when they describe mechanics. The New Mexico Highlands University review of AI case studies reports that KLM used its BlueBot conversational tool to work through a weather-driven backlog of Facebook messages and speeded up customer interaction by 50 per cent, that DHL's warehouse bots sort more than 1,000 small parcels an hour at 99 per cent accuracy, and that DHL reports productivity gains of between 30 and 180 per cent in warehouses that specialise in picking. Read those three as descriptions of a changed process. Each one hands a decision to a machine at a defined point, which is the part most internal proposals leave out.
Does AI adoption pay for itself?
The productivity claim and the revenue claim are two different claims, and only one of them remains valid after an audit. IBM reports that 93 per cent of executives say AI sovereignty must be factored into their 2026 business strategy, and that 53 per cent expect AI to transform business models in their industry by 2030. IBM's own framing of the same research is that companies are seeing productivity benefits while many still struggle to connect AI use to financial outcomes and long-term competitive advantage.
The British Standards Institution's knowledge portal reports that 81 per cent of business leaders say their organisation is already investing in artificial intelligence, and records that in 2022 35 per cent of companies reported using AI in their business while a further 42 per cent said they were exploring it. BSI Knowledge lists the reasons organisations stall before they scale: perceived complexity, data privacy and security concerns, and a lack of trust in the quality and reliability of AI systems. When you write the business case, state the decision that changes, the person responsible for it, and the cost of being wrong in each direction. A productivity percentage provides no information about whether the money returns.
Who is responsible for the decision when the model is wrong?
The accountable person is named on a document that an auditor can test. BS EN ISO/IEC 42001 specifies the requirements for establishing, implementing, maintaining and continually improving an AI management system inside an organisation. BSI Knowledge is the first international standard to provide best practice for governing AI effectively, which experts from fifty countries including the UK developed and structured around the elements any management system needs: senior management support, training, governance processes and risk management.
In practice that creates a register entry per deployment, and the entry has seven fields. The system's name and version. The decision it changes. The person who has that decision. The system processes the data. The action it may take on its own. The review date. The retirement trigger. An entry missing the proprietor field fails the first audit question, and an entry missing the retirement trigger is an authorisation with no end date.
Should you build it, buy it, or hire the work out?
Each of the three routes leaves you holding different obligations, and the differences are visible in what you can point at on day 90.
| Route | What you hold at day 90 | Who repairs it when it drifts | Time to first release | What you own at the end |
|---|---|---|---|---|
| In-house build | A model, a pipeline and the source of every error | Your own team, permanently | The slowest of the three | The code and the operating burden |
| Commercial product | A licence and a configuration | The vendor, on the vendor's release cycle | The fastest of the three | A configuration you can export, and a renewal date |
| Consultancy engagement | A document and a running pilot | Whoever the contract names | The middle case | A pilot that stops when the engagement stops unless you name an owner |
The rows matter because the obligations differ. An in-house build makes you the manufacturer of the model, which brings product liability reasoning into scope. A commercial product makes you the operator, so the vendor controls the model while you control the input data and the permissions. A consultancy engagement makes you the client, and the deliverable is often a report rather than a running service, which means no work continues after the final invoice unless a person inside your organisation takes it over.
When does an AI system get retired?
Retirement happens on a dated condition you agree before go-live. The trigger specifies the date the system is reviewed, the metric that must be met, and the person who decommissions it. A review date on its own leaves the system running by default, because no person is asked to make a decision on that date.
The four documents this page generates all point at that trigger. The definition paragraph decides what falls inside the policy. The inventory gives each system an owner and a decision. The sourcing table tells you which obligations you carry for the next one. The retirement trigger closes the loop so the estate does not grow by accident and every authorisation has a route out of the building.


