A good shop assistant can tell what you're after before you've finished your sentence. They aren't reading your mind. They've simply seen hundreds of customers with the same look on their face, heading down the same aisle.
Predictive interface design tries to give a website some of that instinct. Small, lightweight models notice what a visitor is doing in their current visit and adjust the page to help, without building a profile of who they are.
More than "personalisation"
Most "personalisation" online means being shown things you've already bought, or chased round the internet by adverts for something you searched for once. Predictive interfaces work differently. They look at what someone is doing right now, such as how far they scroll, what they click and how long they spend on a page, to judge what they're trying to do rather than what they did last month.
Say a visitor reads three articles about automating enquiries, moves to your pricing page, then lingers near the Contact button. That's someone getting close to a decision. The page could show a relevant case study, offer a chat, or set the contact form up for the service they were reading about.
The kinds of signal involved
- Which topics a visitor reads, and how deeply
- Whether they look like they're browsing, comparing or ready to decide
- Where they hesitate, go back or give up
- Whether they're on a phone or a computer
- How long the visit lasts and whether they've come back
Designed around privacy
The approach we use doesn't rely on cross-site tracking or on building a profile that follows people around the web. The model works on the current visit only, either in the browser or on your own server. It doesn't need names or email addresses, and nothing is passed to third parties.
When the visit ends, the model forgets it. That makes data protection much simpler, though you still need to describe it honestly in your privacy and cookie notices. Customers get a more helpful site, you get more enquiries, and the data stays on your domain.
"The best prediction engine is the one that respects the user enough not to remember them."
Small models, real results
None of this needs a data science team or racks of graphics cards. Decision trees, small neural networks and even plain rules can predict intent well once they're tuned to how your own visitors behave. We build them in as part of the normal site build, set up for your business.
For a client, that might mean a contact form that changes depending on which service the visitor read about, a shop that reorders its suggestions as someone browses, or a portfolio that puts the most relevant examples first for each visitor's industry.
How we do it
We plan predictive features into a site from the start rather than adding them later. They're built and tested alongside everything else, using the same AI-assisted process that keeps our code clean, our pages quick and our prices fixed.
Want a site that responds to its visitors?
Every package is built the same way, with fixed prices agreed before we start.
View Our Packages →Daedalus Design builds professional websites with AI assistants for UK businesses, and writes about where automation is actually useful in web design.