AI integration means putting an AI step inside software you already run — reading documents into records, answering questions from your own data, sorting and drafting — rather than asking staff to paste things into a chat window. Here is what that costs, and where the money actually goes.
How many UK businesses use AI
The ONS reported in July 2026 that around 35% of UK businesses with ten or more employees use at least one AI technology, up from about 12% in September 2023. Among the smallest businesses (0–9 employees) it was 28%, and among those with 250 or more, 49%. Text generation with large language models is the most common use.
The barrier the ONS heard most was not cost but a lack of expertise — around 18% of businesses reporting obstacles. Cost was named by 7% to 14% depending on size.
The three tiers
ProfileTree's guide to AI in SMEs sets out three levels. Their figures, not ours:
| Level | Typical cost | What it is |
|---|---|---|
| Entry | £250–£2,500 | Off-the-shelf AI tools, configured |
| Mid | £3,000–£15,000 setup, plus subscriptions | AI added to a platform you already use |
| Advanced | £15,000–£50,000+ | Custom-built AI inside your own systems |
Start at the bottom. If an off-the-shelf tool does the job, it is the cheapest answer, and a custom build only earns its cost when the AI has to work inside your own systems, on your own data, with your own approval steps.
What the AI itself costs
Less than most people expect. Mecanik's June 2026 guide for UK SMEs estimates around £0.02 to summarise a 2,000-word document, roughly £20 a month for 1,000 summaries, and £10–£50 a month in API fees at typical small-business volumes.
Those numbers move with the model you choose and how much text you send, so measure them on your own documents during a pilot rather than trusting anyone's estimate — including this one.
Where the money goes
- Getting the data ready. AI is only as good as what it reads. Scanned PDFs, inconsistent spreadsheets and documents spread across inboxes need gathering and cleaning first.
- Connecting it. Reading from and writing to the systems you run — a CRM, accounting, a shared drive — is ordinary integration work, and usually the largest share of a custom build.
- The human checkpoint. A screen where a person approves, corrects or rejects what the AI produced. It is what makes AI safe to use on real work, and it takes design time.
- Logging and limits. A record of what the AI was asked and what it returned, and a cap on spend, so a fault cannot run up a bill.
- Specialist time. Softomate puts UK AI and machine-learning specialists at £800–£1,200 a day, against £400–£550 for a mid-level developer.
Start with a pilot
The cheapest way to find out whether AI helps is to try it on one process, with real documents, for a few weeks. Mecanik makes the point well: learning something for £500 of developer time beats a six-month project scoped before anyone knew whether the idea worked. Pick the process with the most repetitive reading or typing, measure how long it takes now, and measure again.
Sources
- Artificial intelligence in UK businesses: 2023 to 2026 — Office for National Statistics, 20 July 2026
- AI in SMEs — ProfileTree
- AI Integration for UK SMEs: A Practical Guide for 2026 — Mecanik, 26 June 2026
- The True Cost of a Custom Web Application for UK Businesses — Softomate Solutions, updated 1 October 2026