Most small businesses now use AI for something, whether it's drafting emails, answering website enquiries or sorting the books. Whether it is paying off is a different question. In 2025 researchers at MIT looked at company pilots of generative AI and concluded that around 95% had made no measurable difference to profit and loss.
That doesn't mean the tools don't work. It mostly means people adopted them quickly and never set up a way to check. If you pay for AI tools and can't say whether they cover their cost, you're guessing, and guessing gets expensive.
Why measuring beats simply adopting
The big consultancies keep finding the same gap. BCG's 2025 research put the share of companies getting substantial value from AI at scale at about 5%. Most are still at the experimenting stage, spending money without much to show for it.
A small business is actually better placed here. You don't have a sprawling programme to untangle. You have a monthly software bill and a limited number of hours in the week, which makes the sums much simpler. You just need to track the right five things.
Plenty of businesses took up AI because it felt like the thing to do, with no clear idea of what it was for. That's a fine place to start. Six months in, though, you want evidence.
1. Hours saved each week
The easiest one to measure, and usually the one owners care about first. A 2025 Business.com survey found that people using AI at work reckoned it saved them about 5.6 hours a week, with owners and managers reporting more.
Five or six hours is most of a working day. At £50 an hour, that's around £250 a week, or more than £12,000 a year, before you count what you do with the time instead.
How to track it: keep a simple log for two weeks. Each time an AI tool does something you used to do by hand, note the job and roughly how long it would have taken. After a fortnight you'll have a real figure rather than a feeling.
2. Cost per task, before and after
Routine jobs show the biggest difference. Answering the same customer question by hand costs you several minutes of paid time each time it comes up. An assistant answering it costs pennies.
Writing is similar. A blog post or a set of service descriptions might cost a few hundred pounds from a freelancer. A first draft from an AI tool that someone then edits properly costs a fraction of that, though it is never free, because the editing still takes time.
How to track it: pick three jobs you regularly use AI for. Note what each cost before, in money or hours, and what it costs now. The gap is your saving per task.
3. Revenue since you started
Vendor surveys are upbeat here. In Salesforce's research on small and medium-sized businesses, around nine in ten of those using AI said it had increased their revenue.
Those figures are self-reported and come from companies selling AI, so treat them with care. Businesses that are already growing may simply take up new tools faster. It's probably a bit of both.
How to track it: compare the six months before you brought in AI tools with the six months after, allowing for seasonal swings. If revenue rose and AI was the only big change, that is a reasonable signal.
4. How fast customers get an answer
An assistant answers at 2am, on bank holidays and while you're at lunch. For service businesses, speed of reply and winning the job are closely linked. Someone who waits four hours for an answer is far less likely to book than someone who gets a helpful reply in two minutes.
Customers are more comfortable with this than you might expect. Zendesk's customer experience research found that around half of consumers prefer a bot when they want an immediate answer to a simple question.
How to track it: work out your average time to first reply over the past month and compare it with the period before. Most chat tools and CRMs record this for you.
5. What you've stopped paying for
People forget this one. AI doesn't only speed up jobs you still do. It can replace things you used to pay someone else for.
A Claude or ChatGPT subscription costs roughly £20 a month each. For some businesses that has replaced a chunk of the money they used to spend on freelance writing, simple design jobs or admin help. For others it hasn't replaced anything, and that's worth knowing too.
How to track it: list every outside service or contractor you paid a year ago. Mark the ones you've dropped or cut back since using AI, and add up the difference. It's often the most surprising figure of the five.
AI takes over tasks, not judgement. The businesses that get the most from it let it handle the repetitive work and keep a person on the decisions that matter. Automating a bad process just gets you bad results faster.
Putting it together
You don't need a spreadsheet with fifty columns. Check these five once a quarter: hours saved, cost per task, revenue since adoption, response time and outside costs cut.
If they are moving the right way, the spend is justified. If they're flat or going backwards, you're probably using the wrong tool, using it for the wrong job, or not using it well. That's a reason to change your approach rather than give up.
The few businesses getting real value from AI aren't doing anything clever. They picked specific problems, measured what happened and adjusted when the numbers told them to. The technology is the easy part. Most people fall down on the measuring.
Not sure where your AI spend is going?
We help UK businesses build websites and systems where every tool earns its place. No black boxes, no vanity features, just things that work and can be measured.
Get a Free Audit