UK AI Adoption Has Doubled to 32%. Half of Those Firms Spent Nothing. Build the Business Case Before You Spend.

The government has published its second national AI adoption survey, and the headline is a good one: 32% of UK businesses now use at least one AI technology, double the 16% recorded a year earlier. The detail underneath is more useful than the headline, because it shows where the value is real, where it is thin, and why so many firms are still holding back.

If you run a small or mid-sized business and are deciding whether, or how, to invest in AI automation for business in the UK, this article walks through what the survey actually says and turns it into a business case you can test in a few weeks. It builds on our earlier pieces on generic versus bespoke AI and measuring AI return.

What the survey found

The AI adoption survey report 2026 was commissioned by the Department for Science, Innovation and Technology and published on 29 October 2026. It is based on 3,503 telephone interviews with UK businesses carried out between 7 January and 21 April 2026, plus 50 in-depth interviews and seven site visits. The figures that matter most for a decision-maker are these:

Adoption is uneven. Large firms reach 52% uptake, and information and communication leads the sectors at 61%, ahead of real estate at 45% and business services and administration at 39%. The report’s summary does not give a separate figure for small firms in this wave. The previous wave, reported in February 2026, put micro businesses at 14% and medium-sized businesses at 23% against 36% for large firms, so the size gap has been the consistent pattern.

Three things the numbers say about value

1. Most spending is free, and most use is administrative

Nearly half of firms that could estimate their spend reported none in 2025, which the report puts down to free tools and AI features bundled into software people already pay for. Administration is the most common use at 81% of users and planners, followed by marketing at 71% and IT functions at 66%. In other words, a large share of the 32% is staff drafting, summarising and searching with tools already on the desk. That is useful, but it is a long way from a redesigned process, and it explains why only 14% call AI important to their operations.

2. Gains are real but modest, and cost savings are rare

Half of adopters report a productivity increase and 71% report some improvement to workforce productivity, but fewer than a quarter have seen operating costs fall and nearly two-thirds have seen no immediate financial effect. Time saved on tasks does not automatically become money saved. It becomes money only when you either do more with the same people or stop paying for something. Our AI ROI framework covers how to convert hours into pounds honestly.

3. The barriers are skills and a missing business case, not tool prices

Among non-adopters, 65% cited the lack of a clear business case or proven return on investment. Limited skills, expertise or understanding of use cases was cited by 67% of adopters and 78% of businesses planning to adopt. The report also notes that cost was mainly linked to the time, effort and resources needed to implement AI, not to the price of the tools. The constraint is knowing which process to point AI at and having people who can run it well.

Trust is conditional, too. The report finds AI is widely accepted for low-risk work such as drafting and summarising, and less so for high-stakes decisions or autonomous agents. That matches what we see: the safest first projects keep a person in the loop and affect an internal process before they touch a customer.

Why “no business case” is a fixable problem

A missing business case usually means the question has been asked at the wrong level. “Should we adopt AI?” cannot be answered with a number. “Should we automate the way we triage supplier invoices?” can. A business case is simply a before-and-after for one process, with an honest cost on the right-hand side. Here is a method that fits a firm of 10 to 200 people and takes two to three weeks to complete.

A five-step business case for one process

1. Pick one repetitive, rules-based process. Good candidates share three traits: high volume, predictable structure and a clear owner. Inbox triage, document processing, quote preparation and routine reporting usually qualify. Our five processes to automate first is a good starting list.
2. Baseline it for two weeks. Record how many times it happens, how long each takes, who does it, and how often it goes wrong. Use real timings rather than estimates, because estimates are routinely out by a factor of two.
3. Estimate the saving conservatively. Multiply volume by time saved per item, then apply an adoption rate of 60 to 70% rather than 100%. Value the hours at the loaded cost of the people doing the work, and say whether the hours will be redeployed or whether you are avoiding a hire.
4. Count the full cost. Include licences or API usage, set-up and integration time, staff training and the time spent on human review. The survey shows 81% of firms keep some oversight, so budget for it rather than treating it as a failure.
5. Run a time-boxed pilot with a pass mark. Four to six weeks, one team, and a threshold agreed beforehand, for example “handling time down by at least 30% with no rise in errors”. If it misses, you have spent weeks and not months, and you know why.

Free tools first, bespoke when the case demands it

The survey shows firms planning to spend mainly on off-the-shelf applications (61%), task automation (57%) and embedding AI into existing systems (54%). That is a sensible order. Start with what you already own: if you are on Microsoft 365, check what is included before buying more, and make sure your data is in order first, as our oversharing guide explains. Our Microsoft 365 Copilot consultancy page sets out how we approach that.

Move to a bespoke build when the process crosses systems, needs your own business rules, must leave an audit trail, or runs at a volume where per-seat pricing stops making sense. For those cases a purpose-built tool on an API gives you control over data handling and cost. The trade-offs are in our Copilot versus custom AI guide and our integration roadmap.

Keep people and policy in the loop

With 81% of firms keeping human oversight and trust falling away for high-stakes decisions, the sensible default is to design review into the process from day one. Pair it with a short AI usage policy so staff know what data may go where, and find out what is already in use with a shadow AI check. A pilot is much easier to defend to a customer or auditor when the guardrails are written down.

FAQ

Sources: GOV.UK, AI adoption survey report 2026: executive summary (29 October 2026); Resultsense, UK business AI use doubles to 32% (2 November 2026); UKAuthority, DSIT study finds AI adoption is growing (17 February 2026, previous wave). The illustrative calculation is not drawn from a client.