While I agree about not being personal, I don't agree that it has to be data-driven in the formal way you seem to be implying. Its also quite possible to build

While I agree about not being personal, I don't agree that it has to be data-driven in the formal way you seem to be implying. Its also quite possible to build horrendous UX (even it it's supposed to be a dark pattern) while being data-driven, just like it's possible to build garbage software despite having thousands of unit tests with 100% coverage. I think it really comes down to: can your users effectively use the software. To me, having to refer to documentation, go through training, or just "get used to it" are all indicators of poor UX. At the same time, in some cases those are acceptable requirements because, for example, you're building medical UIs that only highly-trained medical professionals are going to use. (I'd still argue you should make it better: a pro won't care that a tooltip defines an acronym, or a warning appears if you set possibly incompatible parameters, etc) What constitutes "good" UX is always going to be somewhat subjective. There's some software that's got a truly awful interface (ever look at ERP stuff?) but whose users spend most of their day using it and are massively effective because they've memorized shortcuts and all the quirks. It has good UX for existing users, awful UX for new users - so how does that get classified?