Where to start with AI if you run a small business

If you run a business with 10 to 100 people, you've probably had the same thought a dozen times this year: "We should be doing something with AI." Then you open your laptop, read three articles that contradict each other, and close it again.

I get it. When Mark and I started working with SMEs on AI, the number one thing we heard wasn't "we don't believe in it." It was "we don't know where to start." That's a completely reasonable position, and it's exactly what this post is for.

Forget the hype. Start with the pain.

The worst way to start with AI is to pick a tool and look for a problem to solve with it. That's backwards. Instead, start with the problems you already have.

Grab a notebook. Think about the last month. Where did your team waste time? Where did things fall through the cracks? What tasks make your best people groan?

Common answers we hear:

"We spend hours every week chasing the same information from the same people"
"Someone has to manually type data from one system into another"
"Our quoting process takes three days when it should take three hours"
"We can't keep on top of follow-ups and leads go cold"

These aren't fancy AI use cases. They're real business problems. And most of them can be meaningfully improved with straightforward automation that doesn't require a computer science degree.

The three-question filter

Once you've got a list of pain points, run each one through these three questions:

1. Is it repetitive?

AI and automation work best on tasks that happen the same way, over and over. If your team does something once a year, it's probably not worth automating. If they do it 50 times a week, that's a strong candidate.

2. Does it follow a pattern?

Even if a task feels complex, it often follows a predictable set of steps. "Read this email, extract the order number, look it up in the system, update the status, reply to the customer." That's a pattern. Patterns can be automated.

3. What's the cost of getting it wrong?

Start with tasks where mistakes are annoying but not catastrophic. You don't want your first AI project to be something where an error costs you a major client. Pick something where a mistake means someone has to correct a spreadsheet, not where it means a compliance breach.

Pick one thing. Just one.

This is the part where most businesses stall. They try to plan a grand AI strategy covering every department, and the sheer scope of it means nothing happens.

Don't do that. Pick one process. The one that scored highest on your three-question filter. That's your pilot.

A good pilot takes two to four weeks, costs a few hundred pounds, and gives you a clear answer: yes, this works, or no, we need a different approach. Either answer is valuable.

The UK government's guide to AI adoption for SMEs reinforces this point. The businesses making progress are the ones starting small, not the ones waiting for a perfect strategy.

What "starting small" actually looks like

Here are three real examples from businesses we've worked with:

A recruitment firm (25 people) was spending 6 hours a week formatting CVs into their house style before sending them to clients. We set up a simple automation that reads the incoming CV, extracts the key information, and reformats it. The time dropped to about 40 minutes a week, and the output was more consistent.

A property management company (15 people) had someone manually logging maintenance requests from emails into their system. We built a workflow that reads incoming emails, identifies the property, categorises the issue, and creates the ticket automatically. The person who used to do that task now spends their time on things that actually need a human.

An accountancy practice (40 people) wanted help with their initial client enquiry responses. We set up an AI assistant that drafts personalised reply emails based on the enquiry details. A team member reviews and sends each one, but the drafting time went from 15 minutes to about 2 minutes per enquiry.

None of these required new hardware. None of them took months. None of them disrupted the business.

The tools question

People always want to know which AI tool to use. Honestly, the tool matters less than you think at this stage. The important thing is understanding your process first.

That said, a few principles:

It should work alongside your existing systems. If someone tells you that you need to replace your entire tech stack before you can use AI, walk away. Good AI solutions connect to what you already have.
You shouldn't need to become a tech expert. If you can't explain what the AI does in plain English, the solution is too complicated. According to research from Be the Business, the most successful SME technology adoptions are the ones where the business owner stays in control of the decisions.
Start with something you can measure. If the task currently takes 5 hours a week, you'll know pretty quickly whether the automation has made a difference.

What happens if you don't start?

I'm not going to pretend the sky is falling. Your business won't collapse tomorrow if you don't adopt AI this quarter. But the honest truth is that your competitors are starting. Some of them already have.

The businesses that start now, even with something small, build capability. Their teams get comfortable with the technology. They learn what works in their specific context. And when a bigger opportunity comes along, they're ready to move quickly.

The businesses that wait tend to stay waiting. And the gap gets wider.

Your next step

Here's what I'd suggest you do this week:

Write down the three tasks your team complains about most
Run them through the three-question filter above
Pick the one that scores highest
Get a quote for a small pilot (we offer free AI opportunity reports that can help you figure out what's worth doing first)

You don't need a grand plan. You don't need a budget committee. You just need one problem and a willingness to try something different.

Most of our clients see meaningful results within 8 weeks of starting. Not because AI is magic, but because the problems they're solving are real, and the solutions are simpler than they expected.