If you run a small or medium-sized business, it is very easy to start your AI thinking with a tool.
Someone mentions ChatGPT. A supplier adds an AI feature. A member of staff asks whether they can use Copilot. A competitor says they are “using AI”. Suddenly the question in the room becomes: “Which AI should we use?”
It sounds sensible, but it is usually the wrong first question.
The better question is: “What problem are we trying to solve?”
That one change makes AI much more practical. It moves the conversation away from hype, demos and subscription lists, and towards business value.
Why starting with the tool causes problems
Most AI tools are impressive in isolation. They can write, summarise, search, classify, draft, analyse, chat and automate. But that does not mean they are automatically useful in your business.
Starting with the tool often leads to three common problems.
First, the tool gets used where it is easiest rather than where it matters. People try it for writing social posts or summarising documents because those are obvious examples. Sometimes that helps. Often it does not touch the real bottleneck in the business.
Second, the business ends up with scattered usage. One person uses ChatGPT. Another tries a meeting note tool. Someone else signs up for an AI design tool. Nobody is quite sure what data is going where, what good output looks like, or whether the cost is justified.
Third, the enthusiasm fades. The first few demonstrations feel exciting, but if the work has not been connected to a specific business outcome, AI becomes another thing people tried for a week and then forgot.
That is not because AI is useless. It is because the implementation started in the wrong place.
Start with the awkward job
A better starting point is to look for the awkward, repeated, time-consuming jobs inside the business.
Not the most glamorous job. Not the one that sounds most “AI”. The one that quietly eats hours, delays customers, frustrates staff, or stops useful work from happening.
For a professional services firm, that might be preparing first-draft client notes after meetings. For an estate agent, it might be turning property, viewing and local-area information into consistent follow-up emails. For a manufacturer, it might be sorting supplier documents or extracting information from technical sheets. For a local membership organisation, it might be keeping events, news and member information up to date.
Once the problem is clear, the AI question becomes much sharper:
- What information does this task need?
- What does a good output look like?
- Who checks it?
- What must never be sent to an external tool?
- How much time or money would a useful improvement save?
- How will we know if it is working?
Those questions are less exciting than asking which model is best. They are also much more likely to produce a useful result.
A local example: Cheltenham Times
One of my own experiments, Cheltenham Times, is a good example of problem-first thinking.
The question was not “Which AI content tool should I use?” The question was: “Could an AI-assisted workflow help build and maintain a genuinely useful local news and information site for Cheltenham?”
That led to a much more practical set of requirements. The workflow had to find local leads, organise information, create drafts, handle basic WordPress tasks, suggest useful sections, and work within sensible rules on privacy, cost and human oversight.
It also had to be corrected. Early outputs needed steering. Geography rules needed tightening. Some images were wrong. Some stories were too thin to publish. The useful work came from shaping the system around the job, not from assuming the AI would magically know what a good local information service should be.
Over time, the experiment expanded beyond articles into practical local assets such as fuel price pages, property price information, planning applications, clubs and events. That is the important lesson for SMEs: the value was not “AI wrote some content”. The value was using AI to support a useful business process and create assets that people might actually use.
The right tool comes later
This does not mean tools do not matter. They do.
Some businesses need a simple assistant for drafting and summarising. Others need a private or local setup because the data is sensitive. Some need an automation that connects existing systems. Others need an agent-style workflow that can perform a sequence of tasks, ask for approval, and leave a proper audit trail.
The point is that you should choose the tool after you understand the job.
If the task involves confidential client information, your choice of tool is partly a privacy decision. If it involves publishing or customer communication, it is partly a quality-control decision. If it runs every day, it is partly a cost-control decision. If staff will rely on it, it is partly a training and change-management decision.
A generic “best AI tool” list will not answer those questions for your business.
A simple way to find your first AI project
If you are unsure where to start, try this exercise with your team.
For one week, ask people to note down tasks that are:
- repeated often;
- based on information that already exists somewhere;
- time-consuming but not especially strategic;
- easy to check once a draft or recommendation exists;
- frustrating enough that people would welcome help.
At the end of the week, pick three candidates and score each one against four questions:
- Would improvement here save time, reduce delay or improve service?
- Is the input information available and reasonably structured?
- Can a human review the output before it matters?
- Can we measure whether the change worked?
The best first AI project is rarely the flashiest. It is usually the one where the task is clear, the risk is manageable, the output can be checked, and the benefit is easy to see.
That might be a document-summary workflow, a sales follow-up assistant, a customer enquiry triage process, a reporting helper, or a small internal knowledge assistant. None of those sound revolutionary. That is the point. Useful AI implementation is usually built from sensible improvements, not grand gestures.
Measure the result, not the excitement
A final mistake is judging AI by how impressive it looks in a demo.
For an SME, better measures are simpler:
- Did it save staff time?
- Did it reduce back-and-forth?
- Did it improve consistency?
- Did it help customers get answers faster?
- Did it reduce errors?
- Did people keep using it after the novelty wore off?
If you cannot measure the change, even roughly, it is hard to know whether the tool is helping or just adding another layer of activity.
The practical next step
Before you buy another AI subscription, write down one business problem you would genuinely like to improve.
Describe the current process. Note where the delays, risks or repetitive work happen. Decide what a better version would look like. Then ask where AI might help.
That order matters.
Problem first. Process second. Tool third.
If you would like help finding the first sensible AI opportunity in your business, I can help you work through the options, identify the risks, and shape a practical implementation plan. Book a short AI consultancy call, or simply get in touch and ask: “What could AI usefully do in our business?”