Where should a small business start with AI? Not with a product. The businesses that get value from AI almost always start with a job that wastes their time, and only then ask whether AI can help with it. The ones that start from "we should be using AI" tend to end up with an impressive demo and nothing to show for it.
This guide helps you find that first job, check that it is a good fit, and rule out the cases where something simpler would do better.
Key takeaways
- Name the problem in one paragraph before you look at any tool.
- Good first jobs are repetitive, frequent, and easy to check, and mistakes are easy to fix.
- Start behind the scenes, not in front of customers.
- Ask whether a simpler fix, or a feature in software you already pay for, would do.
- Pick one task, not a list.
Why does the problem come first?
Because the problem tells you what to measure. "Use AI in the office" cannot succeed or fail. "Spend less time each week typing supplier invoices into the accounts system, with no more errors than now" can.
A clear problem also makes choosing far easier. Once you know the one thing you want to improve, most of the AI products on the market become irrelevant, and you can judge the rest on a single question: will this move my number?
Write it down in a short paragraph:
- The task: what happens, who does it and how often.
- Where it hurts: time lost, errors, delays, customers waiting.
- The measure: the one thing that should get better, and what it is today.
- The owner: the person who will decide whether it worked.
If you cannot name an owner, park the idea. Ideas nobody owns rarely get finished.
What makes a good first AI use case?
In our view, good first candidates share four traits:
- Repetitive and frequent. The same kind of task, many times a week.
- Based on information you already hold. Emails, documents, spreadsheets, records in your systems.
- Easy to check. You can tell quickly whether the result is right.
- Forgiving. If the AI gets one wrong, someone notices and fixes it before any harm is done.
Tasks that depend on unwritten know-how, empathy or delicate judgement are poor first choices. So are tasks where a mistake could hurt a customer, cost real money or create a legal problem. Those may come later, with more experience and firmer controls.
Where should you look?
Start behind the scenes. Internal work carries less risk than anything a customer sees, and it is where many of the quickest returns are found. Common starting points for small firms include:
| Task | Why it often suits AI | What to watch |
|---|---|---|
| Sorting and routing incoming email | High volume, clear categories | Odd or urgent messages need a person |
| Pulling details out of invoices or forms | Repetitive, easy to check | Check totals and bank details by hand |
| Drafting replies to common questions from approved answers | You own and trust the source text | Someone reads before it is sent |
| Routine reporting and summaries | Saves hours, surfaces hidden problems | Check figures against the source |
| Drafting tenders or questionnaires from past answers | Reuses verified content | Facts and claims need checking |
| Turning recorded calls or meetings into notes and FAQs | Low effort, useful output | Ask people before you record them |
A good way to find your own list is simply to ask your team: "Which task do you do most often that you wish you never had to do again?"
Is AI actually the right tool?
Ask this honestly before you spend anything. Often a simpler answer exists:
- A rule or template in your existing software ("if the email contains an order number, send it to the orders inbox").
- A feature that is already part of a package you pay for. Many accounting, customer record and office tools now include AI features, and switching one on is usually cheaper and easier than buying something new.
- A small change to how the work is done.
The way to decide is to compare. Cost out the plain, non-AI fix, then ask what extra the AI option gives you, and at what price. A small improvement over a cheap fix, at many times the cost, is not a good deal.
For most small businesses, the order to try things in is: use what you already have, then buy an off-the-shelf tool, and only then consider anything built for you.
An illustration: the build that is not needed
Imagine a business about to commission a custom AI system that would take months to build. Before signing, the owners sit down and map what their people actually need, what outcome they want and how likely staff are to use it. That exercise shows that a ready-made tool, paid for monthly, would do the job, and the custom build is dropped.
The lesson is simple: write down the need before you talk to suppliers, and you will often find the answer is smaller and cheaper than the first proposal.
How do you choose between several ideas?
If your team gives you more than one candidate, score each one quickly. Use a sheet like this, giving each a score from 1 to 3:
- How often does the task happen?
- How much time or money does it cost now?
- Is the information it needs already digital and reasonably tidy?
- How easy is it to check whether the AI got it right?
- How small is the harm if it gets one wrong?
- Is there someone who will own it?
Pick the top one and start there. Resist running several at once: you will learn more, and be able to tell what worked, if you do one properly. When it is working, the time and confidence it gives you can fund the next.
What if you have already bought a tool first?
That is very common, and fixable. Work backwards: which of your real, repeated tasks could this tool help with? Write the problem paragraph for that task, measure where you are today, and treat the next few weeks as a proper trial. If no task fits, that is useful to know before the renewal date.
When you are ready to run the trial, our guide to why AI pilots stall shows how to set a clear finish line. If you decide to bring in help, a good adviser will want to talk about your problem before their product. The AI project brief guide helps you put it in writing.