A twenty-person accountancy practice recently ran a line-by-line review of its monthly software subscriptions. It found AI features present in six different tools. Two of those features were switched on. The team didn’t know the other four existed. The question the owner asked me afterwards was a good one: what are other businesses like ours actually spending on this, and is our number in the right place?
That question comes up in almost every first conversation now. The answer tends to be more useful than owners expect.
What are UK companies actually spending on AI?
The headline figure for a UK small business is clearer than you might expect. A BEIS-commissioned survey found average annual AI expenditure sits at around £9,500 for a small business, covering software, hardware, data, and consultancy combined. That places AI spending in roughly the same order of magnitude as a modest software stack rather than a capital project. For medium and large firms the figures jump: £380,000 and £1.6 million respectively.
The UK AI sector has grown at speed. Revenue rose from £14.2 billion in 2023 to £23.9 billion in 2024, with the number of AI firms increasing 58% in a single year. Across Europe, corporate spend on AI tools through business expense platforms nearly doubled in 2025, and 73% of more than 2,500 companies now use AI tools in daily operations. The shift, documented by spend management platform Spendesk, is from free trials and proofs of concept to paid infrastructure and recurring workflow subscriptions.
The typical pattern for an owner-managed firm is several low-cost subscriptions alongside one or two larger commitments. The average European company now uses eight different AI tools. For many smaller UK businesses the number is lower, but the direction is consistent across sectors.
Why does knowing the spend benchmark matter for your business?
Knowing what comparable businesses spend on AI helps calibrate two decisions that are easy to get wrong: whether you are underinvesting and falling behind, or spending on tools your team isn’t actually using. The BEIS average of £9,500 is a useful reference point. The more important signal is what is driving that spend, and whether individual tools are producing results you can measure.
Many owner-managed businesses find on auditing their subscriptions that they’re already inside that range or close to it. Microsoft 365 Copilot add-ons, AI tiers in CRM platforms, and AI-enhanced accounting or helpdesk tools all count towards the total. The category has shifted: the question has moved from whether to spend to what you are getting back.
One counterpoint: if major vendors continue bundling increasingly capable AI into base subscriptions without significant price increases, incremental AI spend may stay flat even as actual use grows. In that case, the audit is about activating features you already pay for, not adding more lines to the bill.
Where does the money go across the five cost buckets?
For an owner-managed services firm, AI spend typically falls into five buckets, and understanding which one you’re drawing from helps you compare like with like. The largest bucket for smaller businesses is off-the-shelf productivity tools: general-purpose AI assistants, copilots, and AI tiers within existing SaaS. Integration, data preparation, and compliance advisory time often emerge as significant second and third items once AI moves beyond basic tools.
Productivity tools and copilots. Microsoft Copilot for 365, ChatGPT Team, and similar tools are licensed per user per month. A firm running five or six power users on paid AI tools can expect to spend around £2,000 to £5,000 per year on this bucket.
Vertical SaaS with AI built in. Many businesses encounter AI through tools they would use anyway: CRM platforms with AI-assisted lead scoring, accounting software with anomaly detection, helpdesk tools with triage. This usually appears as an AI add-on tier, typically £10 to £50 per user per month extra. Further up the supply chain, companies such as Featurespace (fraud detection for HSBC) and Tractable (claims assessment for Aviva) sell AI embedded in systems that owner-managed businesses encounter indirectly through their banks and insurers, without contracting for the AI themselves.
Integration and consultancy. For anything beyond off-the-shelf tools, data preparation and integration costs can match or exceed the licence cost in year one. AI consultancies such as Faculty, which has worked with the BBC and the NHS on applied AI projects, typically start multi-month bespoke engagements in the tens of thousands of pounds. Lighter pilots through local specialists run £5,000 to £30,000.
Cloud and infrastructure. For firms using only SaaS tools, cloud costs are bundled into subscription fees. Running your own models is unusual at under fifty staff and rarely justifies the investment at that scale.
Compliance and governance. For regulated sectors, AI use triggers real advisory costs: Data Protection Impact Assessments, vendor due diligence, and policy updates. These appear as extra hours from existing legal or compliance advisers rather than a separate budget line.
When should you spend more, and when is staying lean the right call?
The right level of AI spend depends less on what peers are spending and more on whether individual tools are producing results you can point to. A three-tier budget frame helps: baseline tools for power users, function-specific experiments in one or two departments, and custom integration only where the business case is clear. Review every six to twelve months and cut anything that hasn’t moved a measurable metric.
At baseline, three to five licences for a general-purpose AI assistant run £2,000 to £5,000 per year. Expanding into one or two functions such as AI-assisted email outreach or document drafting adds roughly £5,000 to £15,000 depending on uptake. Custom integration, connecting AI to a case management system or bespoke workflow, starts at around £5,000 and can run to £30,000 with a specialist consultancy.
The moment to spend more is when you can name a specific bottleneck: report first drafts taking too long, client response times that a targeted tool could compress, or a manual process that would take a fraction of the time with the right assistant in place. The moment to stay lean is when you cannot name the before-and-after metric, or when tools are sitting unused because the adoption case was not made properly to the team.
Before adding anything new to the bill, the most useful starting point is a spend map: list every subscription that includes AI features, confirm which ones are switched on, and compare what you pay against what the team actually uses. The gap between paying for and using is where the quick wins are.
What regulatory costs sit inside any UK AI budget?
UK regulators have clear expectations about AI, and the costs of meeting those expectations are real even if they rarely show up as a clean budget line. The Information Commissioner’s Office, the Financial Conduct Authority, and the National Cyber Security Centre all have published guidance that applies to owner-managed businesses using AI, and failing to account for it is a planning error that tends to surface at the worst possible time.
The ICO requires Data Protection Impact Assessments for high-risk AI uses involving personal data, including profiling and automated decision-making. FCA-regulated businesses face additional scrutiny: the FCA’s outsourcing guidance requires documented vendor due diligence, access rights in contracts, and attention to concentration risk when relying on a small number of major AI providers.
The NCSC recommends treating AI systems as part of the broader security perimeter, with restrictions on what data goes to public AI services, logging, and access controls. For a smaller firm, this typically means updating policies and running a brief security review when AI use expands.
For businesses operating across borders, the EU AI Act is shaping vendor pricing even for UK firms. High-risk AI systems sold into the EU face documentation and human oversight requirements, and vendors are building those costs into their pricing.
A practical baseline for a five to fifty person professional services business: budget a few days of adviser time per year for AI governance, with more if your sector has explicit FCA or equivalent expectations. The CMA is also watching how major AI platforms compete, so pricing and availability may shift as scrutiny increases.
The £9,500 annual average gives you a peer reference. The more useful number is the one you arrive at after auditing your own subscriptions, checking usage, and making regulatory costs visible before they become a surprise.
