Most businesses do not need a dramatic AI strategy on day one. They need a useful answer to a simpler question:
What job could AI do here if we treated it like a digital employee rather than a clever toy?
That is the lesson I keep coming back to from the Cheltenham Times experiment.
Cheltenham Times started as a practical test: could an AI agent help build and run a local news and information website for Cheltenham? Not just write the occasional article, but handle a real working brief: find local leads, draft posts, categorise content, suggest useful sections, generate images, help with WordPress workflows, and improve the system over time.
It was not perfect. That is precisely why it has been useful.
For SMEs and professional businesses in Gloucestershire, the interesting part is not local news. The interesting part is the management model. The AI worked best when it had a clear job, the right tools, training, and human oversight.
The problem with vague AI use
A lot of businesses start with AI by asking broad questions:
- Can we use ChatGPT?
- Should we automate something?
- Could AI save us time?
- What are our competitors doing?
Those are understandable questions, but they are not yet a working brief.
If you hired a new member of staff and said, “Please use your initiative and generally improve the business,” you would not expect reliable results. You would give them a role, explain what good looks like, show them the systems, set boundaries, review their work, and correct them when they misunderstand something.
AI agents need the same kind of management.
The mistake is assuming that because AI is fast, it can skip the ordinary disciplines of delegation, process design and quality control. It cannot. It may produce output quickly, but speed does not remove the need for judgement.
The Cheltenham Times job description
The Cheltenham Times agent had a practical job: help create and maintain a useful local information site.
That involved several types of work. It could look for local story leads, check public sources, draft articles, generate featured images, place content into WordPress, suggest new sections, and help maintain useful pages such as fuel prices, property information, planning applications, school catchment information, clubs and events.
That is a broader role than “write a blog post”. It is closer to a junior digital assistant working across research, content, admin and product ideas.
But the job still needed definition. The agent had to learn what counted as Cheltenham, what belonged on a Gloucestershire-wide site instead, which sources were useful, when a story was too thin, and when an image was misleading.
In the case study, a few mistakes led directly to better rules. A generic-looking image for a property story showed the need for image checks. A story that belonged elsewhere led to clearer geography routing. Social posting was moved towards review queues until the process could be trusted.
That is a very normal business lesson: when someone makes a mistake, you improve the process.
Tools matter as much as prompts
A digital employee is only useful if it can use the right tools.
For Cheltenham Times, the useful work did not come from a single prompt in a chat window. It came from connecting the AI to a workflow: source checking, drafting, WordPress, image generation, update routines, and human review.
That is where many AI projects either become useful or fall apart.
If AI sits outside the business, staff have to copy and paste information in and out of it. That can still help, but it is limited. When AI is designed into a workflow, it can become part of how work gets done.
For a solicitor, that might mean preparing first-draft client summaries from approved notes, with a human checking everything before it leaves the firm.
For an estate agency, it might mean monitoring local planning changes, property market information and competitor listings, then creating a weekly review for the team.
For a manufacturer or distributor, it might mean watching supplier price changes, checking product pages, and flagging issues before customers complain.
For a consultancy, it might mean turning call notes, proposals, follow-up actions and research into a more consistent client delivery process.
The tools should fit the job. The AI should not be bolted on for novelty.
Training is business training, not magic
When people hear “training an AI”, they often imagine complex model training or technical work. In many SME use cases, the more important training is ordinary business training.
You show the system examples. You correct bad output. You define unacceptable shortcuts. You build checklists. You decide what should be drafted, what should be reviewed, and what should never be automated.
With Cheltenham Times, the agent improved because mistakes were turned into rules. That is the part business owners should pay attention to.
A useful AI workflow is not usually built in one heroic afternoon. It improves through controlled use. You start with a clear task, review the output, tighten the rules, and expand only when the process earns trust.
That also helps with risk. Instead of giving AI unrestricted access to important systems, you can begin with low-risk work: research, summaries, draft documents, internal reports, update checks or review queues.
Oversight is not optional
The phrase “digital employee” should not mean “unsupervised employee”.
Cheltenham Times still needs human judgement. The AI can scan, suggest, draft and organise, but it is not a magic truth machine. It can misunderstand geography. It can choose the wrong image. It can produce something that looks plausible but needs checking.
That is not a reason to avoid AI. It is a reason to design oversight properly.
In a small business, sensible oversight might include:
- keeping AI-generated customer-facing work in draft until reviewed
- limiting access to sensitive data
- logging what the system has done
- using approved sources rather than the open web for important work
- setting clear “do not do this” rules
- measuring whether the workflow actually saves time or improves quality
The aim is not to remove people from the process. The aim is to remove avoidable busywork so people can spend more time on judgement, service and decisions.
What could this mean for your business?
The Cheltenham Times experiment is not a recommendation that every business should build a news site. Most should not.
The business lesson is broader: AI becomes more useful when you stop treating it as a novelty and start treating it as a managed role.
Ask these questions:
- What recurring job is valuable but awkward, repetitive or often delayed?
- What information does that job need?
- What tools would the AI need access to?
- What does good output look like?
- What must a human check before anything reaches a customer?
- How will we know if it is saving time, reducing errors or creating value?
That is a much better starting point than asking whether your business should “use AI”.
You are not looking for a magic employee. You are looking for a narrow, useful digital role that can be trained, supervised and improved.
A sensible next step
If you run a business in Cheltenham or Gloucestershire and you are wondering where AI could genuinely help, start with one job. Not a whole transformation programme. One job.
Choose something specific enough to describe, frequent enough to matter, and safe enough to test with proper oversight.
If you would like help identifying that first useful AI role, I can help you map the opportunity, design the workflow, set sensible guardrails and work out how to measure whether it is worth continuing.
Book a short AI consultancy call, or simply ask: what useful job could an AI digital employee do in my business every week?