Nobody really wants "AI". They want a result: faster replies, less manual work, more sales. Sometimes AI is the tool that gets you there. Just as often, a plain, well-built feature does the job better — and you don't pay the running costs of AI to keep it alive. So before we add anything clever to a website, we ask one blunt question: what chore or what question are we actually removing?
the question behind the question
People don't want a technology, they want an outcome. "Add AI" is rarely the real goal; "stop answering the same email forty times a week" is. Once you name the outcome, the right tool often turns out to be something simpler — a better form, a clearer page, a smarter search box — and occasionally it really is AI. Starting from the outcome keeps you honest about which. It also stops you from buying a solution before you have defined the problem, which is the most expensive order to do things in.
when ai genuinely earns its place
- Search that copes with messy questions. A large catalogue where people don't know the exact word for the thing they want is a genuinely good use of AI.
- Support that handles the same questions at 2am. If ten questions make up most of your inbox, letting a well-fed assistant field them frees you for the ones that need a human.
- Personalisation at real scale. Hundreds of products and thousands of visitors with genuinely different needs — not a homepage that guesses.
- Turning a pile of content into something findable. Transcribing, tagging and summarising at a volume you would never do by hand.
what it looks like when it's right
A picture helps. Say you run an online shop with two thousand products and a support inbox full of "does this fit a 2018 model?" type questions. A good use of AI here is quiet and specific: a search box that understands a fuzzy human description and returns the right three items, and an assistant that answers the fit questions from your own product data — handing over to a person the moment it isn't sure. Nobody sees a flashy "AI" badge. They just notice they found the thing and got an answer. That is the tell: when AI is doing real work, it tends to disappear into the experience rather than announce itself.
when it's just a sticker
- A chatbot strapped to a five-page site nobody had questions about.
- "AI-powered" in the headline with nothing behind it.
- A recommendation engine for twelve products. (Just show the twelve.)
These help no one. They slow the page down, add something to maintain, and quietly tell visitors you are chasing a trend rather than solving their problem.
the real cost of bolting it on
AI features are not free, and not only in money. They cost a little on every request, add a moment of latency, and need monitoring — because a model can be confidently wrong. A feature that gives a wrong answer five percent of the time in private is an annoyance; the same feature on your public homepage is a liability. That trade is worth making when the upside is real, and not worth making for a badge.
questions to ask anyone selling you ai
If a vendor or agency is pushing an AI feature, three questions cut through the fog fast: What specific task does this remove, and how will we measure that it worked? What happens when it gets something wrong in front of a customer? And what does it cost to run every month, not just to build? If the answers are vague, the feature is probably there for the brochure, not for you.
our rule
We add AI when it removes a real chore or answers a real question — and then we measure whether it actually did. If a humble dropdown does the job, you get a dropdown. Boring beats buzzword, every time. It is the same honest "do you actually need it?" test we apply to whether you need an app or just a good website: start from the job, not the hype.
The best AI on your site is the kind your visitors never notice. It just makes things quicker. If it is there to impress, it is there for the wrong reason. 👋
