Free resource · no email required
The Small Business AI Readiness Checklist
A free 20-minute self-assessment for UK small businesses. Find where AI could help, and whether your foundations are ready first.
- Who it's for
- Owners and managers of UK small businesses who suspect AI is relevant and have no idea where to start
- Format
- Web page, printable on A4, no email address required
- Length
- About five pages when printed on A4, roughly 2,300 words
How to use this
Set aside 20 minutes and find a pen. There are no right answers, and nobody sees this but you. Work through it alone first. Then go through it again with whoever knows the day-to-day work best, who is often not the owner.
A tick means yes. Leave a box empty for no, and write “partly” beside it where that is the honest answer. Print the page on A4 or keep a sheet of paper beside the screen.
The result is a shortlist of one to three areas worth looking into, not a score. If Section A goes badly, stop there. The useful next step is fixing the basics before you look at AI at all.
This is general guidance, not legal or financial advice.
Section A: Foundations
Tick the statements that are true today, not the ones that will be true once you get round to them.
- Our customer details live in one place that someone maintains. Any tool that drafts for customers or talks to them needs to know who they are, and it cannot reconcile three address books and a mobile’s contact list. If no: choose the one place (your job system, accounting software or a single spreadsheet) and move everything else into it.
- We could list our five most common customer enquiries from memory. If you cannot name them, you cannot judge whether a tool is handling the right ones. If no: for two weeks, write one line for every enquiry on a sheet by the phone or in a shared note. The top five will be obvious.
- Someone in the business owns the customer records, by name. Records without an owner drift out of date, and anything built on them repeats the errors with confidence. If no: name a person and give them 15 minutes a week to keep the records tidy.
- Our prices and standard terms are written down somewhere findable. A tool can only repeat what exists in writing. Prices that live in your head stay there. If no: write a one-page price list and half a page of standard terms, even if most jobs are still priced individually.
- We could say how long it takes from enquiry to quote going out. Without a starting figure you will never know whether a change made any difference. If no: note the enquiry date and the quote date for your next 10 jobs. A rough average is enough.
- We could explain our main process to a new starter without doing it in person. A process that exists only in one person’s head cannot be handed to anyone else, human or software. If no: have that person talk it through while someone writes the steps down as a numbered list.
Count your “no” answers, treating each “partly” as half. Three or more means your first useful project is tidying up, and AI can wait until that is done. That is a real result. Fixing these six things makes the business easier to run whether or not you ever use an AI tool.
Section B: Where your week goes
Most owners are wrong about where their time goes, in both directions. Some jobs feel huge and take little time. Others are small and constant, and quietly add up to days.
| Activity | How often | How long each time | Who does it |
|---|---|---|---|
| Answering enquiries (phone, email, messages) | |||
| Quoting | |||
| Scheduling and rebooking | |||
| Chasing customers, suppliers and answers | |||
| Invoicing and chasing payment | |||
| Admin and filing | |||
| Marketing (posts, website, reviews) |
- Estimate in ranges, such as “three to five a day” or “10 to 20 minutes”. A range you believe beats a precise number you invented.
- Include the interruptions. A two-minute call that costs 10 minutes getting back into a job counts as 12 minutes.
- Count the chasing wherever it happens. It hides inside almost every other row.
- When you have finished, circle the two rows that take the most hours in a month and the one you most dislike. Take those into Section C.
Section C: Opportunity signals
Tick each statement that is true of your business in most weeks. Each one names the kind of AI help it points towards and one caveat. A tick means worth looking at, not worth buying.
- The same questions come in over and over. Points towards drafting assistance: a tool drafts replies to common questions from answers you have written, and a person sends them. Caveat: someone still has to catch the exceptions, such as the question that sounds routine and isn’t.
- We miss enquiries out of hours or when we are on a job. Points towards an automated acknowledgement by text or email, so the customer knows they have been heard and when to expect a reply. Caveat: acknowledge, do not answer. A tool that promises prices or dates while you are under a sink will promise the wrong ones.
- Someone retypes the same information into a second system. Points towards data extraction, pulling details out of emails, forms or PDFs into the place they need to go. Caveat: check every figure that matters, especially amounts, dates, part numbers and account details.
- We find out about problems from the customer. Points towards summarising and alerting, such as a daily note of overdue jobs and unanswered messages. Caveat: it only sees what is written down. A problem that lives in someone’s head stays invisible.
- Quotes take days to go out. Points towards drafting a quote from site notes, photos or a call summary, for a person to check. Caveat: pricing decisions stay with a person. The tool can lay out the quote, but it should not decide what the job is worth.
- We write the same emails repeatedly. Points towards templates first, then drafting assistance for the variations. Caveat: your voice matters more than the minutes saved. Customers notice when replies stop sounding like you.
- Old documents and emails are hard to search. Points towards document search, where you ask a question in plain English and get an answer drawn from your own files. Caveat: stale documents give confident wrong answers. Clear out old price lists and superseded procedures first.
- Nobody has time to write marketing. Points towards drafting posts, service pages or newsletters from notes about real jobs. Caveat: generic output reads badly and people recognise it. The raw material has to be your own work and your own opinions.
- Monthly figures take an afternoon to pull together. Points towards summarising exports from your accounting or job software into a short written overview. Caveat: summarising is not bookkeeping. Do not let an AI tool calculate VAT, and keep your accountant or bookkeeper responsible for the numbers.
- We lose track of follow-ups. Points towards reminder workflows that prompt a person when a quote goes unanswered or a customer is due a call. Caveat: a fixed routine and a shared calendar may be the whole answer, with no AI involved.
If you ticked more than five, do not try to address them all. Keep the two or three that overlap with the rows you circled in Section B. That is your shortlist.
Section D: The readiness gate
Before anything goes in front of a customer, you should be able to answer all five of these in a sentence. “We’ll work it out” does not count.
- Would you be comfortable if a customer saw exactly what you sent to an AI tool? If not, you are sending something you should not. Health details are special category data under UK GDPR, so a clinic should take advice before sending any of it.
- Who approves what goes out to a customer? Write down a person’s name. “The system” is not a name.
- What happens when it gets something wrong, who notices, and how? At some point it will. Decide now how a mistake is spotted, corrected and apologised for.
- Is there anyone in the business who could keep this working in six months? Tools change, prices change and people leave. If only one person understands the set-up, write down how it works.
- Do you know where your customer data would be stored, and who else can see it? Check the provider’s terms for where data is held, whether it is used to train their models and how you delete it. If you cannot find out, choose a different tool.
If you cannot tick all five, keep AI away from customers for now. You can still try it on internal tasks using made-up or non-personal information.
Section E: Reading your result
There is no pass mark. Find the pattern that fits you best.
Three or more “no” answers in Section A
Your first project is tidying up, and it pays for itself without any AI.
- Pick the “no” that causes the most daily friction and fix it this month.
- Use your Section B table to choose which fix saves the most time, and do that one next.
- Come back in three months and do Section A again.
Not yet: do not buy an AI tool to cover up messy records. It will only produce the mess faster.
Foundations in place, two or fewer signals in Section C
Start with one narrow, repetitive task, ideally one that customers never see.
- Choose a single task, such as drafting replies to your most common enquiry, and use a general assistant for it.
- Have a person read every output before it is used, for four weeks.
- Compare the time it takes against your Section B estimate, then keep it, change it or drop it.
Not yet: do not connect anything to your customer records or let it send messages on its own.
Foundations in place, three or more signals in Section C
You probably have enough for a small project, possibly an agent that carries out several steps with a person approving the result.
- Group your ticks. Signals that touch the same part of the week, often enquiries, quotes and follow-ups, are one project.
- Map that process step by step before looking at any tool. The Business Workflow Audit, in preparation on this site, will cover this in detail.
- Pass Section D in full, then start with that one project and measure it against your Section B figures.
Not yet: do not sign a long contract or pay a large set-up fee before a trial on your own real work.
What to do next
- Write your shortlist. One to three areas, one sentence each, at the top of this page.
- Fix the Section A answers that matter most. Put a named person and a date against each one.
- Measure one thing now. Enquiry-to-quote time is a good choice, so you can tell later whether a change helped.
- Try one narrow task for four weeks. A person checks every output, and at the end you decide whether to keep, change or drop it.
- Do nothing for a quarter. If cash is tight, the team is stretched or the timing is wrong, waiting three months and doing this checklist again is a sound decision. The tools will still be there.
Plain-English glossary
AI (artificial intelligence). Software that produces text, answers or decisions that would normally need a person, by recognising patterns learned from very large amounts of data.
Assistant. A chat-style AI tool, such as ChatGPT, Claude or Microsoft Copilot, where you type a request and it replies.
Model. The trained system underneath an assistant that actually writes the replies, and different companies make models that vary in quality and price.
Prompt. The instruction or question you give an AI tool, and a clear, specific prompt gets a more useful reply than a vague one.
Agent. An AI tool that works through a sequence of steps towards a goal, such as reading an enquiry, checking the diary and drafting a reply, instead of answering one question.
Automation. Software that follows fixed rules you set, such as “when a form arrives, send this email”, often with no AI involved at all.
Retrieval. Letting an AI tool look things up in your own documents before it answers, so the reply is based on your information instead of general knowledge.
Hallucination. When an AI tool states something false with complete confidence, such as a price, a date or a policy that does not exist.
Token cost. The way many AI services charge, by the amount of text going in and out, which is tiny per request and can add up when a tool runs all day.
About this checklist
Version 1.0, September 2026. Review due March 2027.
Written by Admir Rusidovic, who runs Manchester PC, a computer repair and IT support business in Manchester, and founded Step Into AI.
This is general guidance for UK small businesses. It is not legal, financial or tax advice. For data protection, the Information Commissioner’s Office publishes guidance for small organisations at ico.org.uk. For tax and accounts, speak to your accountant.
This page collects no personal data. There is no form and no sign-up, and no email address is needed. Print it, copy it or share the link.
The outcome
A written shortlist of one to three areas worth investigating, a named starting point from one of three patterns, and either a decision to proceed or a clear, evidenced decision to wait, plus a list of foundations to fix first if Section A failed.
Related
Where this connects
Why AI should fit the business, not the other way around
Most AI projects ask a business to change shape for a tool. Three tests to apply first, and what the national adoption figures suggest.
Why I'm building an AI platform for service businesses
Why I am building Step Into AI for UK service businesses: the missed-call problem behind it, what it is and is not yet, and why no results are published.