A lot of businesses start their AI journey with the same question:
“What tool should we use?”
It is an understandable question. There are new AI products appearing every week, and many of them promise to save time, reduce admin, improve sales, write content, answer customers, analyse documents, or automate half the business.
But in most cases, the tool is not the best place to start.
Before I recommend any AI tool, workflow or agent, I want to understand how the business actually works. Not the version on the website. The real version: the repeated admin, the awkward handovers, the spreadsheets that only one person understands, the emails that keep being rewritten, the bottlenecks everyone has learned to work around.
That is what an AI audit is for. It is not a big corporate strategy exercise. For a small business, it should be a practical look at where AI could help, where it would be a distraction, and what needs to be in place before anything useful is built.
1. Where is work being repeated?
The first thing I look for is repetition.
Not every repeated task should be automated, but repeated work is often where AI can help quickly. Examples include:
- answering similar customer enquiries
- turning notes into follow-up emails
- summarising long documents
- preparing quotes or proposals from a standard pattern
- drafting social posts from existing material
- checking incoming leads and routing them to the right person
- updating records after calls or meetings
The useful question is not “can AI do this?” It is “does this happen often enough, and predictably enough, to be worth improving?”
A task that takes five minutes once a month is probably not worth building around. A task that takes 20 minutes every day, across several people, starts to look more interesting.
2. Where are the bottlenecks?
AI is often more useful when it supports a bottleneck than when it simply makes an already-fast task slightly faster.
In many SMEs, the bottleneck is not obvious at first. It might be the owner approving everything. It might be a senior team member who holds too much process knowledge in their head. It might be customer emails waiting for someone to decide what to say. It might be sales opportunities getting stuck because nobody has time to follow up properly.
This matters because a good AI project should remove pressure from the system, not just add another tool for people to check.
For example, an AI assistant might prepare first-draft replies, summarise a customer history before a call, or flag which enquiries need urgent attention. The human still decides. But the thinking time is reduced, and the bottleneck starts to loosen.
3. What data and knowledge already exist?
AI is only as useful as the information it can safely and reliably use.
So I look at where the business knowledge lives:
- website pages
- brochures and PDFs
- price lists
- CRM records
- call notes
- spreadsheets
- support emails
- policy documents
- project folders
- shared drives
- the owner’s head
That last one is common.
Many small businesses have plenty of knowledge, but it is scattered. Before building anything clever, it may be better to tidy the source material, define the current process, or create a simple knowledge base.
This is also where privacy comes in. Some information is fine to use with general AI tools. Some should not leave the business. Some can be used only if access, permissions and retention are properly considered. A sensible AI audit separates those categories before anyone starts copying customer data into a chatbot.
4. What software is already in use?
Most businesses do not need another disconnected app.
They already have email, calendars, spreadsheets, accounting software, a CRM, a website, shared storage, messaging tools and industry-specific systems. The question is whether AI can fit around those tools sensibly.
Sometimes the best answer is very simple: use AI to draft better emails, summarise meetings, or turn messy notes into structured tasks.
Sometimes it is more integrated: an AI workflow that reads a form submission, checks existing records, drafts a response and creates a task for a human to review.
My own Cheltenham Times experiment is a useful proof point here. The value did not come from one magic AI writing tool. It came from connecting a real objective with sources, WordPress, rules, local knowledge, review points and repeatable workflows. The AI could scan, sort, summarise and suggest, but it needed clear boundaries and human judgement.
That same principle applies in a professional services firm, a local retailer, a consultancy, a trades business or a growing B2B company. AI is more useful when it is part of the workflow, not sitting off to the side as a novelty.
5. What skills and habits does the team have?
A good AI recommendation has to match the people who will use it.
Some teams are confident experimenting. Others are cautious, busy, or rightly concerned about mistakes. Some staff will use a tool well after ten minutes of guidance. Others need examples, templates and clear rules.
This is not a weakness. It is normal implementation reality.
If a business has never used AI before, I would rather start with a small, safe, visible use case than a complicated agent that touches important systems. The first win might be a shared prompt library, a simple meeting-summary workflow, or a customer enquiry triage process with human approval.
Once people see what good looks like, you can build from there.
6. Where are the risks?
This is the part that gets missed when AI is sold as a shortcut.
Every audit should ask:
- Could this expose private customer or staff data?
- Could it produce advice that needs professional review?
- Could a mistake harm the business’s reputation?
- Could the system take action when it should only make a suggestion?
- Who checks the output?
- What happens when it is wrong?
These questions do not mean “do nothing”. They mean design the workflow properly.
For many SMEs, the right approach is not full automation. It is assisted work: AI prepares, summarises, drafts, checks or flags; a person reviews and decides. That is often where the best return sits, especially early on.
What a useful AI audit should produce
At the end of an AI audit, you should not be left with a shopping list of random tools.
You should have a clear view of:
- the best first use cases
- which tasks are worth improving
- what data or documents are needed
- what risks need guardrails
- what can be done quickly
- what should wait
- how success will be measured
For some businesses, the answer might be a simple staff assistant. For others, it might be a workflow that reduces admin in sales, customer service, marketing or operations. Occasionally, the best recommendation is to fix the process first and add AI later.
That honesty matters. AI should earn its place in the business.
If you run a business in Gloucestershire or Cheltenham and you are wondering what AI could sensibly do for you, start with an audit rather than a tool. Look for the repeated work, the bottlenecks, the data, the software, the skills and the risk areas.
Or, if you would like an outside view, get in touch and book a short AI consultancy call. We can look at how your business actually works and identify the first practical AI opportunities worth testing.
