Where its predecessor only looked at the first interaction, INP considers the worst of them, which makes it a much harsher and much more honest metric. It is almost always a JavaScript problem: long tasks blocking the main thread, oversized event handlers, hydration of components nobody scrolled to, and third-party tags — consent managers, tag managers, chat widgets, analytics — competing for the same thread.
Fixing it is organisationally harder than fixing LCP because the offending scripts frequently belong to marketing rather than engineering. Break long tasks up, defer what is not needed for the first interaction, and audit the tag manager: on many sites the single biggest INP win is removing tags nobody has looked at in two years.