Where to start with AI: pick one problem, not a platform
2 October 2026 ยท 2 min read
There's a lot of pressure at the moment to "do something with AI". It can make it feel like you need a big plan, a new platform and a serious budget before you're allowed to start.
You don't. The businesses that get real value from AI usually start with something small and slightly boring. They fix it, see that it works, and then do the next one.
Start with the work nobody enjoys
Ask your team what they'd happily never do again. You'll usually get answers fast. The good candidates tend to have a few things in common:
- It's done the same way every time
- Someone is copying information from one place into another
- It involves chasing people, or reminding them
- Mistakes creep in because it's dull and people rush it
- It quietly piles up when that one person is on leave
Then be a bit ruthless about it
For each task on the list, think about three things. Roughly how many hours does it take each week? What does it cost you when it's late or goes wrong? And are the rules clear, or does it need a lot of judgement? The best place to start is something that takes real time, hurts when it slips, and follows clear rules.
Keep the first project small
If you can't explain it in one sentence, it's probably too big for a first go. Something like "when a new enquiry comes in on WhatsApp, reply and add it to our CRM" is about right. Small enough to build properly, visible enough that your team notices the difference.
You probably don't need new software
Most businesses already have the tools they need. The problem is that they don't talk to each other, so people end up doing the joining-up by hand. Connecting what you already have, and letting AI handle the steps in between, is often where the quickest wins are.
Write down the before
Before you start, note how long the task takes today. A few weeks after it's live, check again. It sounds obvious, but it's the easiest way to know whether it actually helped, and it makes the case for whatever you automate next.