Staying ahead
Where to start with AI in a small business
You know you should be doing something with AI. You open ChatGPT, type a question, get a decent answer, and then go back to the job you were already late on.
Are you late to AI?
No. Most UK businesses are still early, and the ones using AI mostly use it for one or two things. Starting now with one well-chosen job puts you in a better place than a business that bought five tools and set none of them up.
What the official figures say
The ONS report on artificial intelligence in UK businesses from 2023 to 2026 found around 35% of businesses with 10 or more employees used at least one AI technology by June 2026, up from around 12% in late 2023. In construction it was 13%. The ONS called adoption relatively shallow: the average firm using AI uses fewer than two kinds of it.
The most common first use is the dull one
The same ONS report found improving business operations was the most common use of AI, reported by over 60% of larger businesses. That means the admin: replies, paperwork, summaries and sorting. Not robots, and not replacing staff.
- 35%of businesses with 10+ staff using any AI, June 2026
- 12%the same figure in late 2023
- 13%of construction businesses using AI
Source ONS, Artificial intelligence in UK businesses: 2023 to 2026 (published 20 July 2026).
What makes a good first AI job?
A good first AI job happens often, follows a pattern, and fails safely. If you could brief a new starter on it in ten minutes, and a wrong answer would be spotted before it reached a customer, it is a strong candidate. The aim of the first job is to learn how AI behaves in your business, cheaply.
It comes round every week
A job you do twice a year won't teach you much, and won't save much either. A café owner answering the same booking questions every day, or a plumber writing up job notes every evening, has a job that repays the set-up quickly.
It follows a pattern you can write down
If you can describe what a good answer looks like, the AI can be told the same thing. A bookkeeper chasing late invoices follows a pattern: polite first reminder, firmer second, a call after that. That is a job with rules, and rules are what AI follows best.
A mistake is cheap and easy to spot
Your first job should be one where you see the output before anyone else does. A drafted reply that sits in your outbox until you press send is safe. A message that goes straight to a customer is not, at least not yet.
You already know what good looks like
Pick a job you do well yourself. You will spot a weak answer at a glance, and you can write down what it got wrong. Handing AI a job you don't understand only means you can't tell when it is wrong.
How to score your candidate jobs in 15 minutes
- 01
List five jobs that eat your week
Not "marketing" or "admin", but specific jobs: replying to quote requests, writing up job notes, chasing unpaid invoices, posting the weekly specials, answering the same five questions about parking. Next to each, guess the hours it takes in a normal week.
- 02
Give each job a point for each test it passes
One point each for: it happens every week, you could write the rules on one page, you see the output before a customer does, and you do it well yourself. A job needs all four points to be a first job. Three points makes it a second job.
- 03
Pick the four-pointer with the most hours
As an example, a plumber might score enquiry replies at four points and 3 hours a week, job notes at four points and 2 hours, and invoice chasing at three points because a wrong reminder to a good customer is costly (example figures). Enquiry replies win.
- 04
Write down what success looks like
Before you start, decide what would make it worth keeping: say, half the drafts sent unchanged by week four, and no wrong facts sent to a customer. Our guide to trialling AI on one job explains how to set that bar.
Good first AI jobs, by type of business
Every business is different, but the same kinds of job come up again and again. These are common first jobs that pass all four tests for most firms of each type.
Trades: plumbers, electricians, builders
Drafting replies to quote requests from your price list and the areas you cover. Turning voice notes from the van into tidy job notes. Writing the follow-up email after a job with the guarantee details. All three are drafts you read before they go.
Shops and cafés
Answering repeat questions about opening hours, allergens, parking and click-and-collect from a fact sheet you wrote. Turning the specials board and a photo into the week's social posts for you to approve. Summarising supplier emails into a list of what needs ordering.
Clinics and practices
Drafting replies to appointment and fee questions that contain no clinical details. Tidying policy documents and patient leaflets. Health information is special category data under UK GDPR, so keep the AI away from clinical records at the start, and read the ICO's guidance on data protection impact assessments before any job that touches them. The GDPR Compliance Kit sorts the data map and records that go with it.
Agencies and offices
Turning a client call transcript into a brief. Drafting the weekly status email from your project notes. Producing first drafts of proposals from a template and past examples. If the job is paperwork that follows a set format, such as invoices and contracts, Document Automation is built for it.
Which jobs should you leave until later?
Leave anything that moves money, deletes things or needs judgement about people until your first job has run well for a while. These aren't off limits forever. They are where a mistake costs the most, so they come after you trust the set-up. Our post on which AI jobs still need a person to sign off covers where the approval step stays.
Anything that pays or refunds
Paying a supplier or issuing a refund should stay with a person. AI can prepare the payment run or the refund note, and you approve it. The approval step costs you seconds and protects you from the expensive mistake.
Anything that deletes or sends on its own
Deleting old records, cancelling bookings or sending messages without a check are all hard to undo. Start with drafts, and keep sending behind a person until the job has a track record. When a job has earned it, Autonomous Agent Operations runs repeat jobs overnight behind a safety lock and leaves you a morning report of what it did.
Hard conversations
A complaint from a regular, a dispute with a venue, a difficult message to staff. AI can suggest a first draft, but the decision and the tone are yours. Customers can usually tell when they are talking to a template.
Your first month with AI, week by week
- 01
Week 1: write the job down
Describe the job as you would to a new starter: what comes in, what good looks like, what to never say, and two or three real examples of each. Tell it to say when it doesn't know rather than guess; Anthropic's guide to reducing hallucinations puts that first. This page becomes the AI's standing instructions, and how to write AI instructions it will follow has a template.
- 02
Week 2: set it up as drafts only
Give the AI only the access the job needs. For enquiry replies, that means reading the enquiry and writing a draft. It does not need to send, see your bank or delete anything. Our post on giving AI only the tools it needs goes into the detail.
- 03
Week 3: compare it with your own work
Check every draft before it goes. Note each one you had to change and why. Each reason is a missing rule, so add it to the instructions and the same mistake gets rarer. That habit has its own post: every AI mistake should become a rule.
- 04
Week 4: decide with evidence
Count how many drafts went out unchanged and how many needed fixing. If the job is saving you time and the fixes are shrinking, keep it and consider a second job. If not, change the instructions or pick a different job.
Common mistakes when starting with AI
Most false starts come from starting too big, not from choosing the wrong tool.
Starting with the tool instead of the job
Buying a subscription and then looking for something to do with it leads to a month of interesting chats and no time saved. Pick the job first. The tool is usually the one next to your email, as our post on which AI model a small business should use explains.
Using a personal free account for customer details
The quickest way to start is the riskiest place to put customer emails. Set the job up on an account the business owns. Our post on free or paid AI for business covers what changes, and the ICO's advice for small organisations covers your data protection basics.
Skipping the dull security basics
An AI account is another account that can be broken into. Use the same care as for your email. The NCSC's small organisations guide to cyber security covers backups and protecting your devices and accounts, much of it in a few minutes per step.
Judging it on week one
The first week's drafts are the worst they will ever be, because the instructions are new. Owners who stop after three bad drafts never see week four. Give it the full month, fixing the instructions as you go.
Questions
08Asked and answered
Drafting replies to common enquiries is the easiest first AI job for most small businesses, because it happens daily and you check each draft before it goes. You already know the right answers, so you spot a weak one at a glance. If paperwork eats more of your week, see Document Automation.
The AI tool you already have access to is usually fine for a first job, because the job matters more than the tool. If you run on Google Workspace, try Gemini; on Microsoft 365, try Copilot. We compare the main options in which AI model a small business should use.
Count how many outputs you use unchanged, and whether the fixes shrink week by week. If by week four most drafts go out with light edits and no wrong facts, it is working. If the same mistakes keep coming back, fix the instructions first. Our guide to trialling AI on one job sets a pass or fail bar.
Give AI only what the one job needs, read-only where possible, and never the ability to pay or delete. For enquiry replies, that means reading enquiries and writing drafts, not sending them. Widen access one step at a time, once the job has a record. See giving AI only the tools it needs.
It can be, if the AI sees only what the job needs and the data stays inside accounts your business owns, on a plan that does not train on your data. Avoid personal free accounts for customer details. The ICO's AI guidance covers your duties, and our safety rules set out what stays off unless you switch it on.
Most owners can tell within a month whether a first AI job is saving time, if they count drafts used and edits made each week. Week one is slow while the instructions settle; the saving shows in weeks three and four. Keep notes as you go, as set out in every AI mistake should become a rule.
Yes: we set up AI for owner-run businesses at fixed prices, with no sales call. You pick the job, we build it inside accounts you own, and it keeps working with a spend cap and an off switch. Every price is on the prices page, and how it works shows the three steps.
This is general guidance, not legal or financial advice. For your own situation, speak to a qualified adviser.