We hear a lot of big claims about AI. It will replace jobs, transform industries, reinvent everything. But when you’re actually running a business, the real questions are much simpler:
- Where does AI genuinely help in day-to-day operations?
- How do I use it safely, without creating more chaos or risk?
- What’s realistic for an SME with limited time and budget?
In a recent session [hosted by the ABM] with fellow mentors and business owners, we explored exactly that: practical ways to use AI to improve operational efficiency, not as a buzzword, but as a working tool.
This blog pulls together the key ideas and turns them into concrete steps you can apply with your team or your mentees.
AI isn’t going away but it is already in your business
A recent survey from the CIPD suggested that 1 in 6 employers expect job cuts due to AI in the next year. At the same time, a CIO in a multi-national organisation told me recently they are adamant they’ll use AI to reshuffle roles rather than cut them.
Whatever your view, one thing is clear: AI use is already happening in your business, whether you’ve “adopted AI” or not.
In one MIT study, only about 40% of companies said they were using large language models (like ChatGPT or Claude) – but around 90% of employees were using them anyway in some form.
That gap is important. It means:
You may think “we don’t use AI here”,
But your people are likely already pasting text, documents or even customer information into free tools.
Practical takeaway:
Rather than pretending AI isn’t being used, it’s safer to acknowledge it.
- Set some basic guidelines (what’s okay to paste into AI, what’s not).
- Encourage staff to use approved tools with proper privacy controls, instead of random free websites.
Before you automate, fix the process first
One of the most important lines I shared in the session was from a friend who works in AI and cybersecurity: “Automation without design only speeds up inefficiency.”
If your process is clunky, confusing, or unclear, AI will happily help you do the wrong things faster.
Practical takeaway: Before you throw AI at a problem, ask:
- What is the process today?
- Where are the real bottlenecks?
- Is AI actually needed – or would a simple, “boring” automation be enough?
I often encourage businesses and mentees to map their operations visually: from customer enquiry → delivery → invoicing → aftercare.
Once it’s mapped, then the questions become:
- Where do delays happen?
- Where are errors frequent?
- Where do people spend time on repetitive, low-value tasks?
That’s where AI has the best chance of making “the boat go faster”, rather than just adding cost and complexity.
Not all AI is the same: Pick the right type for the job
We talked in the session about three broad types of AI most people encounter:
Chat AI – e.g. ChatGPT, Claude, Gemini
Great for drafting, summarising, brainstorming, rewriting, and basic analysis.
Copilot AI – e.g. Microsoft Copilot inside 365
Works inside tools like Word, Excel, PowerPoint and Outlook.
Good for summarising emails, pulling actions from meetings, analysing spreadsheets.
Agentic AI – tools that can take actions, not just write text
For example, an automation that pulls data every Monday, runs a report, and emails it to your team without you touching it.
These often use tools like Zapier, Make, n8n combined with AI.
Practical takeaway: You don’t need to remember the labels, but you do need to be clear on your intent:
- If you want better wording → Chat AI.
- If you want help inside Office tools → Copilot.
- If you want something to happen automatically every week → you’re into automation/agent territory.
Garbage in, garbage out: how to prompt better
You’ll have heard the phrase ‘garbage in, garbage out’. With AI, it’s painfully accurate.
If you give the AI vague, incomplete input, you’ll get vague, unreliable output. The good news is you don’t need to become a “prompt engineer”, just be a bit more deliberate.
Practical takeaway: I often use a simple prompting framework I call GCSE:
G – Goal:
What are you trying to achieve? (e.g. “Summarise this 10-page report into 5 bullet points for a non-technical director.”)
C – Context:
What background does it need? (e.g. “We’re a B2B service business working with schools.”)
S – Source:
What should it base its answer on? (e.g. paste the document, link, or data, or mention your ideal customer profile.)
E – Expectation:
How should the answer look? (e.g. “Keep it under 200 words, UK English, professional but friendly.”)
You can literally write your prompt like a mini-brief. For example:
“You are an operations consultant. Goal: Help me compare three suppliers for steel sheets.
Context: We’re a UK manufacturing business.
Source: Use the price and delivery details in this table.
Expectation: Give me a short list of pros and cons and a recommendation in under 150 words.”
You’ll almost always get a better result from that than “Which supplier is best?”.
Hallucinations, free tools and data risk
Every AI system sometimes hallucinates meaning it makes things up in a very confident tone.
ChatGPT is powerful but will rarely say “I don’t know”.
Claude is better at admitting uncertainty, but still not perfect.
So, you should:
- Never blindly trust AI on safety-critical or legally sensitive decisions.
- Always double-check facts, especially dates, numbers, and references.
- For research, use tools that show sources (e.g. Perplexity, or ask explicitly: “Show your sources.”).
Then there’s the data question. If a tool is free, you are the product. That doesn’t mean you should never touch free tools, but it does mean:
Don’t paste personal, confidential, or commercially sensitive data into free accounts.
If you’re using AI with client, staff or financial data, use a paid, business-grade setup with clear privacy guarantees.
Practical takeaway: Most major tools also offer temporary or incognito chats, which are useful when you don’t want your data used to train the model.
Easy, practical use cases you can try now
Here are some real operational use cases we discussed that you can adapt:
Onboarding new staff or clients
Use AI to:
- Turn your existing policies and guides into a clear onboarding pack.
- Generate checklists for “Week 1”, “Month 1”, etc.
- Draft welcome emails and explainer documents.
Screening suppliers
If you’re B Corp certified (or aiming to be), you’re under pressure to vet suppliers more thoroughly.
AI can help by:
- Summarising long policy documents.
- Highlighting relevant certifications or red flags.
- Creating side-by-side comparison tables.
You still make the decision; AI just speeds up the reading.
Mapping customers and routes
One real example:
A client had 62 school sites and wanted to understand their geography.
With AI, you can:
- Paste in addresses and ask: “Group these by region and suggest efficient visit clusters.”
- Combine with route planning tools for more efficient scheduling.
Turning paperwork photos into usable data
In one operational business, teams were taking photos of handwritten delivery notes.
We used a simple flow to:
- Drop the photo into a Teams channel.
- Run an AI step to read the handwritten name and key details.
- Convert it to a PDF and save it with a meaningful file name. It’s not glamorous, but it saved hours of manual scanning and renaming.
Weekly operational “heads-up” emails
Using tools like Zapier or Make, you can set up an automation to:
- Pull weather forecasts for a specific area.
- Check road closures or major works (useful if you have vehicles on the road).
- Have AI summarise the impact and email a weekly briefing to transport managers.
Five years ago, that would have been a complex IT project. Now it’s often a few hours’ setup.
Start in the “fast lane” and the “slow lane”
One analogy I love from a continuous improvement engineer is thinking of data as two swimming pool lanes:
The fast lane – low-risk, non-sensitive data where you can experiment and iterate quickly.
The slow lane – sensitive or high-impact areas (customer data, HR, finance) where you move carefully, test thoroughly, and involve IT or legal if needed.
Practical takeaway: When you’re working with mentees or your own team, ask:
- What’s a fast-lane use case we can try this month?
e.g. social media content, internal summaries, basic automation.
- What might be a slow-lane project to explore more carefully?
e.g. anything involving personal data, compliance, or major process changes.
As a leader or mentor your role isn’t to sell AI
Finally, an important point that came up in discussion, we’re not here to sell AI. We’re here to help people use it wisely.
That means:
- Asking more questions than you answer: Where are you now? What’s your readiness? What’s your risk appetite?
- Encouraging experimentation, but also boundaries.
- Helping people learn how to ask better questions of AI, and how to review the answers critically.
AI can absolutely help small and medium-sized businesses to punch above their weight, but only if it’s grounded in good process, clear thinking, and human judgment.
If you’d like to discuss using AI in your business, please get in touch.
