AI saves an owner-led firm the most time on admin that starts with words: first drafts of quotes and emails, summaries of calls, reading long documents and replying to routine enquiries. It saves little where one wrong number or clause is expensive. The working rule is simple. Let it produce a fast first draft, and keep a named person responsible for what goes out.
This guide covers the jobs where that rule pays off today, the jobs where it does not, and a short list of where to start this month. It is written for owners and managers of firms with no IT department, not for people who already live in the tools.
Where UK businesses are with AI
The numbers vary a lot depending on who is asking. The Department for Science, Innovation and Technology (DSIT) interviewed 3,500 businesses between February and May 2025 and found around one in six (16%) were using AI. Among those using or planning to use it, 70% named administrative and support work as a use case.
The British Chambers of Commerce (BCC) surveyed 668 businesses, mostly small and medium-sized, for a report published in March 2026 and found 54% were actively adopting AI. Most were using generic tools such as chatbots and writing assistants, for drafting content, summarising information, analysing data and research. Only around one in ten had built something specific to their own business.
The two surveys ask different questions of different firms, so do not read them as one trend line. What they agree on is the shape of it: in most firms, “using AI” means a person with a chat assistant doing admin faster. That is a perfectly good place to be, and it is the place this guide is about.
The jobs where it saves time today
First drafts of quotes and emails
Give it the customer’s enquiry, your price list and the usual terms, and ask for a reply. You get a structured first draft in seconds instead of starting from a blank page. The saving is the drafting, not the thinking: you still decide the price and the scope.
A worked example. Paste in the customer’s message and your price list, then ask:
Draft a reply to this enquiry. Confirm what they asked for, give the price from the list, say what is not included, and offer two dates for a site visit. Plain English, under 150 words. Use only the prices I have given you.
Before it goes out, check every figure against your price list, check the scope matches what you would actually do, and check the dates are real. It will occasionally state a price that is not in your list, whatever you told it. That is why the check is not optional.
Summarising calls and meetings
Recording or transcribing a call and asking for a summary, decisions and a list of actions is one of the clearest wins. If the notes contain personal data, the ICO’s advice on an AI meeting-notes tool is to update your privacy notice to cover the tool and check the outputs for accuracy. Tell people when a call is being recorded or transcribed, and read the summary before you share it.
Reading documents
Supplier terms, tender documents, insurance schedules, a 40 page specification. Ask it to pull out the payment terms, the deadlines and anything unusual, and to tell you where in the document each one sits. Then go to that page and read it yourself. The tool finds things quickly. You confirm them.
Routine enquiries
Most firms answer the same ten questions repeatedly: opening hours, lead times, what is covered, how to book. Keep your best answers in one document and have AI draft replies from them. A person still reads and sends each one. Speed matters more than most owners expect, which is the point of our piece on response time and conversions. A customer-facing chatbot that answers on its own is a different, larger decision, and the next section explains why.
Spreadsheet formulas
Describe what you want in plain words, such as “add up column D for every row where column B says Approved”, and ask for the formula. It is quick and usually right, and it takes the fear out of spreadsheets that have grown beyond what one person understands. If your workbook has become the system, our note on signs your business has outgrown spreadsheets covers what comes next.
Tidying notes into reports
Site visit notes, a voice memo, a messy bullet list: ask for a clean report with headings and an actions table. This is the same job as drafting, applied to your own notes, and it is where people tend to save the most time without anyone noticing.
The jobs where it does not help yet
Exact numbers
AI is better with words than with arithmetic. It can summarise a table of figures well and still get a total or a margin wrong, sometimes confidently. Ask it to write the formula, then let the spreadsheet do the sum. Anything involving VAT, retentions, payroll or a quote total gets checked by a person against the source, every time.
Legal and contractual wording as the final word
It is good at drafting a first version and at explaining what a clause seems to say. It is not a lawyer, and it may not be answering for UK law. The BCC report makes this exact point: asking ChatGPT what a dismissal process needs to include may produce an answer based on US law. Use it to prepare for a conversation with your solicitor, not to replace one.
A process nobody has written down
If the way you price a job, chase a late payment or approve a purchase exists only in one person’s head, AI has nothing to work from. It will produce something plausible, and plausible is worse than blank. The fix is to write the steps down first, even roughly. That exercise is useful whether or not you ever use AI, and any automation needs it before it starts.
Customer data in free or personal accounts
The tool matters as much as the task. On free and personal plans your data can be used to improve the model by default, while business plans do not train on it by default. Our guide to where your data goes in ChatGPT and Claude sets out the differences. If a task involves customers, use a business account.
Where to start this month
You do not need a strategy for this. You need four steps and a month.
- Pick one person and one task. Someone who already writes a lot of quotes or emails. Choose a task they do at least weekly, so they get many chances to learn it.
- Use a business account. Put the first user on a business plan under the company, not a personal login, so the data protections apply from day one.
- Write down the check. For that task, list what a human must verify before anything leaves: figures, names, dates, scope. Put the list where they will see it.
- Review after four weeks. Ask what saved time, what created rework and what they stopped doing. Keep what worked, drop what did not, and only then add a second task or a second person.
That is deliberately modest. Most of the value in the first few months comes from a handful of repeated jobs done well, not from a broad programme.
When it is worth getting help
The jobs above all involve a person typing into a chat window. The next step is connecting AI to the systems where your quotes, invoices and enquiries actually live, so the first draft arrives without anyone pasting anything. That is engineering work, and it needs careful rules about what the system may do alone. Our AI consultancy page explains how we approach it.
If you would rather talk it through first, book a 30 minute call. We will tell you which two or three of these jobs are worth trying in your business, and which to leave alone for now.