Most AI projects stall because they start with the tool, not the problem. Pick the right use case first, and the rest gets a lot simpler. Here’s the test we use.
“Do something with AI.”
It’s a phrase we hear a lot at the moment, usually from a business owner who knows they should be paying attention but has no obvious place to start. The pressure is understandable. More than half of UK SMEs are now using AI in some form (54%, according to the British Chambers of Commerce), yet only around 16% have deployed it with a clear business purpose behind it. Plenty of businesses are dabbling, but far fewer are actually doing anything that moves the needle.
That gap is where most of the frustration lives, and it nearly always comes down to the same mistake: starting with the tool instead of the problem. Choosing the right AI use case, the first thing genuinely worth your time, is what separates a bit of tinkering from a real result. Here’s how to go about it.
The Trap: Trying to “Do AI” Without a Use Case
Here’s how it usually goes. Someone reads that AI is transforming business, buys a licence or signs up for a tool, and waits for the transformation to just ‘happen’. But ultimately, it doesn’t, and the tool sits there half-used while nobody’s quite sure what it’s for and the whole thing quietly gets shelved.
This is more common than the headlines suggest. Gartner found that around 30% of generative AI projects were abandoned after the proof-of-concept stage by the end of 2025, often because there was no clear business case to scale them. It even has a name now: pilot purgatory.
The other version of this trap is quieter. Staff start using ChatGPT or Copilot off their own bat to save a bit of time or make things easier for themselves, but they do so without anyone deciding what’s allowed or what data is safe to put in. That becomes a risk waiting to surface.
Both problems have the same root cause: starting with the tool instead of the problem. A good AI use case works the other way round. It begins with something that’s already costing you time, money, or sleep and asks whether AI is the sensible way to fix it. That’s the shift that turns AI and automation from a talking point into a result.
What Makes a First AI Use Case Worth Doing?
Not every problem is an AI problem, and not every AI win is worth chasing first. When we help clients weigh this up, five questions do most of the work. If a use case can’t answer them, it’s probably not the one to start with.
It solves a real, repetitive problem: Begin with something that already eats time or causes friction, not with a tool you’ve heard good things about. This sounds obvious, but it’s the single biggest sticking point: the most-cited barrier for UK businesses is simply identifying where AI fits, named by around 39% of firms. Start with the pain, and the use case chooses itself.
The saving is measurable: If you can’t say what “better” looks like, you won’t know whether it worked. It needs to be attached to a quantifiable metric like the amount of hours saved, errors reduced, or even a job that stops landing on someone’s desk. Write down the baseline before you start, so the result isn’t a matter of opinion.
The data it touches is low-risk or properly governed: A use case built on public, low-sensitivity information is a safer place to learn than one wired into client records or financials. Where sensitive data is involved, the guardrails need to come first. This is where cyber security and AI decisions stop being separate conversations.
It works with what you already pay for. The quickest wins are usually sitting inside tools you’re already subscribed to, Microsoft 365 Copilot being the obvious one, switched off or never configured. Before buying anything new, it’s worth checking what your current AI and automation capability can already do.
It’s small enough to prove quickly: One task, one tool, one measurable result. Prove it works, then expand. The businesses that struggle are the ones trying to transform everything at once; the ones that succeed grow from a single proven win.
Run a use case through those five and you’ll quickly separate the ideas worth piloting from the ones that just sound impressive in a meeting.
A Worked Example
Here’s the kind of situation we see regularly. A business of around 30 people comes to us wanting to “use AI somewhere”, without much idea where. Rather than start with a shopping list of tools, we start with a question: what’s the job everyone quietly dreads?
Often, it’s writing things up. After every client meeting, someone spends 20 to 30 minutes turning scribbled notes into a tidy summary and a list of actions. Multiply that across the team and the week, and it adds up to real, recoverable time.
A task like that ticks every box. The problem is real and repetitive, the saving is measurable (potentially a day or a week across the team), the data involved is low-sensitivity internal notes rather than anything confidential, the tool is often already paid for (Microsoft 365 Copilot, sitting unused), and the scope is small enough to prove in a fortnight.
From there, it’s simple. Switch on the right feature, agree on a few clear rules about what can and can’t go into it, and let one team try it for two weeks. When the summaries come back faster, and sometimes tidier, than the manual versions, you’ve got your proof. Only then is it worth looking at where else the same approach could go.
Notice what didn’t happen. Nobody bought a new platform, ran a six-month project, or “transformed” anything. They picked one job worth doing and did it properly.
When to Bring in Your MSP
You can work through the five criteria yourself, and for a simple first use case, you might not need anyone else. But there’s a reason so many businesses stall at this point: the skills to do it confidently often aren’t in the building. Nearly half of small firms say they don’t have the in-house knowledge to use AI well (46%, according to the FSB), and that’s exactly the gap a good managed service provider fills.
An MSP earns its place on the parts you can’t easily see. The security and data check before anything goes live. The audit of tools you’re already paying for, so you’re not buying something you own twice over. And, just as importantly, the honest “this one isn’t worth it” when a use case looks shinier than it is.
That’s the thinking behind our AI / Business MOT: a straightforward review of where AI could genuinely help your business and where it couldn’t. It’s the same practical, plain-English approach our clients have relied on for years, now pointed at the question everyone’s asking.
Come and Talk It Through
Choosing your first AI use case isn’t really a technical decision. It’s a business one: what’s worth your time, what’s safe to try, and what’s just noise. Get that right and the tools look after themselves.
If you’d like to think it through with people who do this every day, we’re hosting an afternoon session that’s built for exactly that. AI Unlocked is a short, practical look at where AI fits for growing businesses, followed by drinks and networking.
It’s on Thursday 18 June 2026 at the South Place Hotel, a few minutes from Liverpool Street. Spaces are limited, and they’re going to the people who’ll get the most from the room.
Register for the event
Come along, bring your questions, and leave with a clearer idea of the one AI project actually worth doing next.
