Most AI projects do not fail because the technology is not clever enough.
They fail because nobody has agreed what useful looks like.
A business owner hears that AI can save time, improve customer service, speed up admin, help with marketing, summarise emails, analyse documents, support sales and automate half the working day. All of that may be true in the right setting. But it is also far too broad to act on.
A sensible AI consultancy engagement should not start with a model, a tool, or a dramatic promise about replacing staff. It should start with the business.
The question is not “how can we use AI?”
The better question is: “Where is work getting stuck, repeated, delayed, missed or made more expensive than it needs to be?”
That is where useful AI projects usually begin.
Step 1: understand how the business actually works
The first stage is listening and mapping.
For a Gloucestershire professional firm, that might mean looking at how enquiries arrive, how quotes are prepared, how client information is handled, how documents are reviewed, how follow-ups happen and where staff lose time.
For a local service business, it might mean checking how bookings are managed, how job notes are recorded, how customer questions are answered, how reviews are requested and how repeat work is encouraged.
This does not need to be a six-month transformation programme. In many small businesses, a few focused conversations and a look at the existing workflow will reveal the main opportunities quickly.
The aim is to understand the work before recommending AI. Otherwise, the result is usually a shiny tool looking for a problem.
Step 2: find practical use cases
Once the business is understood, the next job is to find use cases.
Good use cases are usually specific. Not “use AI for marketing”, but “turn approved job notes into a first-draft case study”. Not “automate customer service”, but “draft replies to common enquiries for a human to approve”. Not “analyse documents”, but “extract key dates, obligations and risks from incoming paperwork”.
The best early use cases often sit in work that is:
- repetitive but still needs judgement
- research-heavy
- admin-heavy
- slow because information is scattered
- valuable but often postponed
- currently done manually with copy and paste
This is also the point where privacy, risk and data access need to be discussed. Some workflows can use external AI tools safely. Others may need tighter controls, redaction, local systems or a human approval step before anything leaves the business.
A good engagement should make those trade-offs explicit, not hide them.
Step 3: prioritise quick wins, not fantasy projects
Most businesses do not need their first AI project to be huge.
They need something useful, contained and measurable.
That means ranking use cases by practical value, difficulty, risk and speed. A simple priority list is often enough:
- What would save staff time quickly?
- What would improve response times or reduce missed follow-ups?
- What would reduce errors?
- What is safe enough to test without disrupting the business?
- What can be measured before and after?
This is where consultancy should be commercially grounded. If a workflow saves ten minutes a week, it may not be worth building. If it saves three hours a week, improves consistency, or stops good leads being missed, it becomes more interesting.
The aim is not to impress people with AI. The aim is to improve a real business process.
Step 4: build and test one workflow
After prioritising, build one workflow properly.
That might be an AI assistant that drafts client follow-up emails from meeting notes. It might be a system that checks a website, inbox or spreadsheet each morning and highlights important changes. It might be an internal research assistant that prepares briefing notes from approved sources. It might be a sales support workflow that helps turn enquiries into next actions.
The important point is that the first version should be treated as a working prototype, not a finished magic employee.
In my Cheltenham Times experiment, AI agents helped research local stories, organise sources, draft content, create images, suggest site sections and maintain useful local pages. But the useful lessons came from managing the mistakes: wrong images, thin source material, geography rules, review queues and clearer publishing checks. Each problem led to a better rule, test or workflow.
That is exactly how business AI should be approached. Start small, watch what happens, tighten the guardrails and improve the system.
Step 5: train staff and set the rules
AI does not remove the need for people. In most sensible SME projects, it changes what people spend time doing.
Staff need to know:
- what the AI is allowed to do
- what it is not allowed to do
- when a human must approve the output
- what data can and cannot be used
- how to spot weak or risky answers
- how to give feedback so the workflow improves
This training does not need to be academic. It should be based on the actual workflow being introduced.
If the AI drafts client emails, staff should practise reviewing and correcting those drafts. If it summarises documents, they should know what must be checked manually. If it prepares lead research, they should know how to verify the source before acting on it.
A short, practical training session around a real workflow is usually more valuable than a generic “AI awareness” presentation.
Step 6: measure and improve
A sensible engagement should end with measurement, not just installation.
Before-and-after measures do not need to be complicated. They might include time saved per task, number of enquiries followed up, response time, error rate, draft quality, staff confidence or customer experience.
In my AI Visibility Audit project, the useful part was not simply that AI was involved. The tool showed businesses what ChatGPT and Perplexity said when asked to recommend someone like them, then gave practical fixes. That is the pattern worth copying: diagnose the issue, produce useful output, and make the next step clear.
The same applies inside a business. If the AI workflow does not produce a clearer next action, save meaningful time, reduce risk or improve consistency, it needs changing.
What this looks like in practice
A good first engagement might look like this:
- a short discovery session to understand the business and pain points
- a simple map of possible AI use cases
- a priority shortlist based on value, risk and effort
- one workflow built and tested with real examples
- practical staff training
- a review of results and next improvements
That is a long way from buying a subscription and hoping staff will find a use for it.
It is also more realistic than pretending AI will transform the business overnight.
For many Cheltenham and Gloucestershire SMEs, the right first step is modest: one useful workflow, with clear rules, human oversight and a way to measure whether it helped.
Once that works, you can decide what to improve next.
Want to know what AI could do in your business?
If you are interested in AI but do not want hype, guesswork or an expensive false start, I can help you identify the most useful first opportunities in your business and build a sensible workflow to test.
Book a short AI consultancy call, or simply get in touch and ask: “what could AI realistically do here?”
