Most of the advice written about artificial intelligence is aimed at businesses with a thousand staff and a data team. This is not that. This is what AI does for a firm of five, fifteen or fifty people in the UK in 2026, written in the order you would actually ask the questions.
The short version: AI is genuinely useful for four jobs, it costs less than one part-time hire and the businesses getting value from it are the ones who picked a single boring task and finished the job rather than the ones who bought a licence for everybody and hoped.
The four things AI reliably does well
Strip away the branding and almost every credible business use of AI in 2026 falls into one of four buckets. If a supplier is pitching you something that does not sit in one of these, ask harder questions.
1. It drafts and rewrites
First drafts of proposals, job adverts, policy documents, product descriptions and awkward emails. AI does not write better than your best writer. It writes faster than a blank page, and it will happily produce the fourth version of a paragraph without sighing. The value is in the time between nothing and something.
2. It summarises
A ninety-minute recorded meeting becomes eight bullet points and a list of who agreed to do what. A forty-page tender document becomes a one-page brief with the deadlines pulled out. This is the use that converts sceptics fastest because the output is checkable in thirty seconds.
3. It extracts structure from mess
Supplier invoices arriving as PDFs in fourteen different layouts. CVs. Delivery notes photographed on a phone. Handwritten job sheets. AI reads the mess and returns consistent fields you can push into a spreadsheet, a finance system or a CRM. For a lot of businesses this is where the largest single time saving hides.
4. It answers questions against your own documents
Point a tool at your policies, price lists, product manuals and past quotes, and staff can ask it questions in plain English instead of hunting through SharePoint. This is what Microsoft is selling when it talks about Copilot understanding your work, and it is only as good as the documents underneath it.
If you cannot describe the task in one sentence without using the word artificial, it is probably not ready to be automated.
What it costs, honestly
Prices moved in 2026 and they moved upwards. Microsoft raised Microsoft 365 business plan pricing from 1 July 2026 and began bundling Copilot into Business Standard and Business Premium, which changes the arithmetic for a lot of UK firms. As a working budget for a ten-person business:
- Microsoft 365 Copilot — roughly £22 per user per month on an annual commitment, with cheaper small-business tiers now available.
- A general assistant such as ChatGPT Business or Claude for Work — around £20 per user per month.
- Automated workflows — a few pounds a month in run charges once built, plus the build itself.
- Setting it up properly — usually the largest line in year one and the one most often left out of the business case.
A sensible starting position is not a licence for everybody. It is three or four licences for the people who write, quote or triage most, run for a quarter, measured honestly.
Where it still gets things wrong
AI is confident in exactly the same tone whether it is right or wrong, and that is the entire risk in one sentence. The failures we see most often in UK businesses are dull rather than dramatic:
- Invented specifics. Prices, dates, part numbers and legislation references that look plausible and are not real.
- Stale knowledge. A model trained months ago quoting a VAT threshold or a Microsoft plan name that has since changed.
- Confident summarising of the wrong document. Particularly where two versions of a contract sit in the same folder.
- Tone that is not yours. Left unchecked, everything drifts towards the same mid-Atlantic corporate voice.
The fix is not clever prompting. It is a rule that a human signs off anything that leaves the business, and that anything with a number in it gets checked against the source.
A first ninety days that actually works
The businesses that get value do roughly this.
Weeks one and two: find the task
Ask every team to name the job they most resent. Look for tasks that recur weekly, take more than an hour and produce text or extract data. Write down how long it takes today. You will need that number later and nobody ever remembers it afterwards.
Weeks three to six: run it small
Three or four people, one task, proper licences on a business plan. Set the ground rules in writing before anyone starts: what may be pasted in, what may not and who checks the output. Keep the old process running alongside for the first fortnight.
Weeks seven to ten: measure
Compare against the number you wrote down. If the saving is under twenty per cent, the task was wrong, not the tool. Pick another one. If it is over fifty per cent, you have found something worth building properly.
Weeks eleven and twelve: make it permanent
Write the process down. Train the people who were not in the pilot. Decide what happens when the person who set it up is on holiday. This is the step that gets skipped, and it is the difference between a saving and an anecdote.
The governance you cannot skip
You do not need a forty-page AI policy. You need one page that answers four questions: which tools are approved, what data may go into them, who checks the output before it goes out and who to tell when something goes wrong. Put it where staff will find it and review it twice a year.
If you handle personal data — and almost every business does — UK GDPR still applies in full. AI is not a special category that suspends it. Your privacy notice needs to reflect the processing you are actually doing, and your supplier due diligence needs to cover the AI vendor the same way it covers your accountant.
Where this leaves you
AI in 2026 is not a competitive advantage on its own. Everyone has access to the same models. The advantage is in choosing the right task, connecting the tool to your actual data and writing down the process so it survives contact with a busy week.
That is ordinary operational work, and it is the part we do. If you want a second opinion on where AI would genuinely pay for itself in your business, our IT consultancy team will walk your processes with you and tell you plainly which ones are worth automating and which are not. You can start a conversation here.
