Before you buy any AI tool, you should be able to answer ten questions: what problem you are solving, how you will know it worked, where the information lives, how clean it is, who can see what, what data must stay private, what your staff already use, which tools you already pay for, who will own it, and how a person will check the results. If you can answer most of them clearly, you are ready to start small. If several draw a blank, fix those first. That usually costs less than an AI tool that never gets used.
This checklist works whether you have 10 staff or 250, and you do not need to be technical to use it. Set aside an hour with whoever runs your day-to-day operations and go through it together.
Goals: what are you actually trying to fix?
1. What specific problem do you want AI to solve?
"We should be using AI" is not a problem. "Our two office staff spend most of Monday typing supplier invoices into the accounting system" is. Good starting points are tasks that are repetitive, text-heavy and follow rules most of the time: reading emails and forms, pulling details out of documents, sorting requests, drafting routine replies, and answering the same internal questions again and again.
Write down two or three candidate tasks, who does them today, and roughly how often. If you cannot name one, the right next step is to look at how work gets done, not to buy software.
2. How will you know if it worked?
Pick one or two simple measures before you start: hours spent per week, turnaround time, number of errors, or how many requests sit unanswered. Write down today's figure, even if it is an honest estimate. Without a starting point, every AI project "seems to help" and nobody can say whether it was worth the money. Our guide to what AI automation costs shows how to weigh that against the price.
Data: is your information ready to be used?
3. Where does the information for this task live?
AI tools are only as useful as the information they can reach. For each candidate task, list where the inputs come from: a shared inbox, a folder of PDFs, a CRM (customer relationship management system), an accounting package such as QuickBooks Online or Xero, or a spreadsheet on someone's desktop.
If the answer is "in Sarah's head" or "it depends who you ask", that is a finding in itself. AI cannot use knowledge that was never written down.
4. Is that information reasonably clean and current?
You do not need perfect data, but you need data you would trust a new employee to work from. Watch for duplicate customer records, three versions of the same price list, out-of-date procedures, and folders nobody has tidied in years. An AI assistant will happily quote an old policy with total confidence.
A useful test: if you gave the same files to a sensible temp, could they do the job? If not, an AI tool will struggle too.
Security: who can see what?
5. Are access permissions set up properly?
Many AI tools, including the AI features built into Microsoft 365 and Google Workspace, can see whatever the signed-in person can see. That is fine if permissions are tidy. It is a problem if salary spreadsheets, HR files or board documents are shared with "everyone in the company" because that was easier at the time.
Before switching on AI that searches your files, check who has access to sensitive folders and sites, and remove sharing that no longer makes sense. Your IT provider can usually help with this.
6. What data must never go into an AI tool?
Decide this before anyone starts. Typical examples are customer personal information, payment and bank details, passwords, health information, and anything covered by a confidentiality agreement. Business-grade AI services usually offer stronger data protections than free consumer versions, but the terms differ by product and plan, so read them or ask someone to check. If you work in a regulated industry, talk to your advisor about what applies to you.
People: who will use it, and who owns it?
7. What are your staff already using?
Ask, without blame. In most businesses, someone is already using a free AI chatbot to draft emails or summarize documents. That tells you two things: where people see value, and where data may already be leaving your control. A short, plain-English policy fixes the second part. Our free AI usage policy template and guide to AI usage policies cover what to include.
8. What tools do you already pay for?
Check whether your existing subscriptions already include AI features, or offer them as an add-on. Your email and office suite, CRM, help desk and accounting software may all have AI options you have not switched on. Using what you already own is often the cheapest and safest first step, because the data and permissions are already in place.
9. Who will own it?
Every AI tool needs a named person who decides how it is used, answers questions, notices when it goes wrong and keeps the setup current. It does not have to be a technical person, but it has to be someone with time set aside. AI projects with no owner tend to drift: a few people use the tool for a few weeks, then everyone goes back to the old way.
10. How will a person check the results?
AI tools make confident mistakes. For each task, decide where a person reviews the output before it reaches a customer, a supplier or your accounts. For low-risk work, such as a first draft of an internal email, a quick read is enough. For anything involving money, contracts or customer commitments, build in a clear review step and keep a record of what the AI did.
How to score yourself
Go through the ten questions and mark each one:
| Answer | What it means |
|---|---|
| Clear yes | You know the answer and could explain it to a colleague |
| Partly | You have a rough idea, but it is not written down or agreed |
| Not yet | Nobody knows, or the honest answer is "it's a mess" |
This is a conversation tool, not a formal score. As a rough guide:
- Mostly clear yes: you are ready for a small, well-defined pilot on one task. Keep it narrow and measure it.
- A mix of yes and partly: pick the one task where you scored best, fix the gaps for that task only, then pilot it.
- Several "not yet" on data or security: sort those out first. Tidying permissions and agreeing a policy are low-cost steps that make every later AI project safer.
- "Not yet" on goals or ownership: pause. Buying tools before you know what you want them to do, or who will look after them, is the most common way AI money gets wasted.
Common mistakes to avoid
- Starting with the tool instead of the problem. A vendor demo is not a business case.
- Trying to do everything at once. One task done well teaches you more than five half-finished pilots.
- Skipping the boring parts. Permissions, policy and clean data are not exciting, but they decide whether AI is safe to use.
- Assuming staff will just pick it up. Short, practical training with real examples from your business makes the difference between a tool people use and one they ignore.
Where to start
If you answered the ten questions and know your next step, go for it: pick one task, set a measure and run a small pilot.
If you would rather have an independent view, our AI readiness assessment works through these questions with your team over one to two weeks. You get a prioritized list of opportunities, a recommended AI usage policy and a 90-day roadmap. When you are ready to build, we can deliver the top items through AI automation and help your team adopt them with adoption and enablement support.
Not sure which fits? Get in touch and tell us what you are trying to fix.
