The limits of adaptive interfaces
By Allan Leone on
Personalisation is the trend of the year and it has a failure mode that rarely makes it onto the slide. When an interface reshapes itself per person, it stops being something two people can talk about.
Adaptive interfaces are genuinely useful and the case for them is easy to make. The case against is more specific and it only shows up after launch, usually in support tickets.
When the interface reshapes itself for each person, it stops being a shared object. That sounds abstract until you list what depends on it being shared.
What breaks when the layout is personal
- Support cannot walk someone through a screen they cannot see. Every call starts with reconstructing what the user is looking at.
- Colleagues cannot give directions. "It is in the third tab" stops being true, and in team products this is constant.
- Documentation describes a layout that may not exist for the reader, and screenshots go stale in a way you cannot detect.
- Muscle memory never forms, because the target keeps moving. For daily-use tools this is a real productivity cost.
A predictable interface is a feature. Every rearrangement spends trust you may not get back.
The distinction that makes it work
The useful split is between payload and furniture.
Payload is content and defaults. What is in the list, what is pre-filled, what is surfaced first, which of forty settings you see at the top. Personalising these is almost always positive and it is invisible in the bad way, which is to say nobody notices it working.
Furniture is structure. Navigation, the location of primary actions, the shape of the page. Personalising these usually costs more than it returns, because the gain is a few seconds of scanning and the loss is every property in the list above.
If you must move the furniture
Sometimes the case is strong enough. When it is, make the change explicit rather than silent. Let the user see that something moved, tell them why, and give them a way back to the default. Silent rearrangement is what produces the feeling that the software is unreliable.
Where to start
- Audit what your product currently personalises and sort each item into payload or furniture. Most teams have not made this distinction explicitly.
- For anything in the furniture column, ask what the support call sounds like. If you cannot answer, that is the answer.
- Add a visible reset to default for any adaptive structure you keep.
- Check whether your own documentation still matches what a new user sees. If it does not, the personalisation is already costing you.
Tags: ux, ai, product