Almost every client conversation we have now includes the words “and we’d like to add AI”. Fair enough. But most businesses don’t need an AI strategy. They need one or two AI features that clearly earn their place.
So we flip the question. Instead of “how can we use AI?”, we ask “which of our existing problems has AI made cheaper, faster or easier to solve?” That usually leads somewhere much more useful.
Where AI tends to pay off
Search. Users struggle to find the right product, article or answer, especially when they don’t know the exact words. Search that understands what someone means, not just what they typed, can make a real difference to sales and to support volume.
Summaries and first drafts. Long documents, support threads and meeting notes are slow to read and slow to write. A good summary or a first draft saves time, and a person still decides what goes out.
Sorting and routing. Tickets, leads and documents often get read and sorted by hand. AI can tag them and send them to the right place, with a person checking the unclear ones.
Pulling out data. Getting structured information out of emails, PDFs, forms and photos is tedious and easy to get wrong. Right now it’s one of the most reliable things AI does.
Personalised content. Tailoring lessons, recommendations or messages to each user used to take a lot of manual work. We’ve seen this first hand in education: LingoTeach builds a full bilingual lesson in about 90 seconds, and Tandem turns a child’s ideas into an illustrated story. In both cases a teacher or parent stays in charge of the result.
Where to be careful
Be wary anywhere a wrong answer is expensive and hard to spot. Money calculations, medical or legal decisions, anything a user will rely on without checking. AI can still help there, but it needs guardrails, a human review step and a clear fallback when it isn’t sure.
How to pick your first use case
For each idea, we ask a few blunt questions:
- How often does this happen? Frequent, repetitive tasks pay back fastest.
- What does it cost today? Hours spent, mistakes made, customers lost.
- What happens when the AI gets it wrong? Can the user see it and fix it easily?
- Is the data there? Can the feature get what it needs, securely?
- How will we know it worked? Time saved, conversion, tickets resolved, happier users.
Pick the one that scores well on all five. Ship it to a small group, measure it, and only then expand.
Building it properly
An AI feature in production needs the same care as any other feature, and a few extra things on top. Personal data has to be handled properly. Costs need to stay predictable as usage grows. Someone needs to keep an eye on the quality of the output. And the product should still behave sensibly when the AI service is slow or down.
Plan for those from day one and you end up with a feature people rely on. Skip them and you usually end up with a nice demo that never quite makes it to launch.
