The question that sends AI projects off course
A lot of AI conversations start in the wrong place.
Someone in the business has seen a demo, tried ChatGPT, heard about Microsoft Copilot, watched a LinkedIn video, or been told by a supplier that an AI tool will save hours every week. The first question then becomes:
Which AI should we use?
It is an understandable question. It is also usually the wrong first question.
For most SMEs and professional firms, the better starting point is:
What problem are we trying to solve?
That sounds obvious, but it changes the whole project. It moves the conversation away from hype and towards business value. It also makes AI safer to introduce, because you can decide where it fits, who checks the output, and how you will know whether it has helped.
Why tool-first AI becomes messy
A tool-first approach usually starts with enthusiasm and ends with confusion.
A business signs up for an AI product, gives a few people access, and waits for the productivity gain. Some people use it daily. Others ignore it. A few use it in risky ways because nobody has set rules. After a month or two, nobody is quite sure whether it has saved time, improved quality, reduced admin, or simply added another subscription to the pile.
The problem is not that the tool is bad. The problem is that the business never defined the job.
But AI is not like buying a printer or switching phone systems. It can affect information, decisions, customer communication and errors. That means it needs a clearer brief.
Start with the business problem
A better approach is to pick one useful, contained problem and work backwards.
For example:
- We spend too long turning meeting notes into follow-up actions.
- Our inbox is full of repetitive enquiries that still need a human response.
- We lose time searching through policies, past proposals or technical documents.
- Our CRM is incomplete because nobody has time to update it properly.
- We want to monitor local opportunities, competitor updates or public information without checking dozens of websites manually.
- We produce reports, summaries or marketing drafts from the same trusted sources every week.
These are much stronger starting points than “we need AI”.
Once the problem is clear, you can ask more useful questions:
- What information does the AI need access to?
- Is that information sensitive?
- Should the output be a draft, a recommendation or an automated action?
- Who reviews it?
- What does good look like?
- How much time, cost or rework should this save?
Only then does it make sense to choose a tool.
A practical example: Cheltenham Times
My Cheltenham Times experiment is a good example of why the job matters more than the tool.
The aim was not “use AI to write articles”. The job was broader and more practical: build a local news and information website for Cheltenham, find useful local stories, organise them, create helpful sections and test what AI agents could do on a real project.
That meant the AI-assisted workflow was judged by whether it created something useful for local readers, not by whether it looked impressive in a demo. It checked local websites, public feeds, council updates, events pages, business news, community sources and social signals. It helped draft posts, categorise content, generate images, monitor sources and suggest improvements.
It also made mistakes. Some images needed better checks. Some geography rules had to be tightened. Some social posting needed to move into review queues. Each issue led to a clearer rule, test or guardrail.
That is the point. The value was not “an AI tool”. The value was a managed workflow with a real purpose, human oversight and room to improve.
The same pattern applies to an accountancy firm, estate agency, consultancy, manufacturer, legal practice or local service business. The useful question is not “which AI platform is best?” It is “what job do we want this system to do, and how will we manage it safely?”
Make it safe, useful and measurable
Before choosing a tool, I would suggest defining three things.
1. Safe
What data is involved? Customer details? Staff information? Contracts? Financials? Medical, legal or commercially sensitive material?
If the work involves sensitive information, the answer may affect the choice of tool, settings, permissions and whether a local or private system is more appropriate. It should also affect what staff are allowed to paste into public tools.
Safety also includes human review. For many SME use cases, AI should draft, sort, summarise or suggest — not make unchecked decisions.
2. Useful
A use case should remove friction from real work. If nobody cares about the problem, the AI project will not last.
Good early projects are often boring on the surface: summarising enquiries, preparing first-draft replies, checking documents, updating CRM notes, turning calls into actions, or creating draft reports from approved sources.
That is fine.
3. Measurable
Decide how you will know whether it worked.
That does not need to be complicated. You might measure hours saved, faster response times, fewer missed follow-ups, reduced rework, or simply whether staff keep using the workflow after the novelty wears off.
If you cannot describe the expected benefit, pause before buying another AI subscription.
A simple way to choose your first AI project
If you are not sure where to start, try this exercise with your team.
Ask everyone to list tasks that are:
- repeated often
- based on information that already exists
- time-consuming but not especially creative
- easy for a human to check
- valuable enough to matter if improved
Then pick one. Not five. One.
Write down the current process, the pain point, the data involved, the risks, the person responsible for review and the result you want. That brief will tell you far more about the right AI approach than a generic list of popular tools.
The tool comes later
AI can be genuinely useful for smaller businesses, but only when it is attached to a real business problem.
Starting with the tool makes the project feel exciting at the beginning and vague later. Starting with the problem makes it easier to build something safe, useful and measurable.
If you run a business in Gloucestershire or Cheltenham and you are wondering what AI could sensibly do in your organisation, I can help you identify practical opportunities, choose a first project and build the right guardrails around it.
Book a short AI consultancy call and we can start with the most important question: what problem are we solving?
