Defining what constitutes "float" in terms accessible to the system would be one challenge - you'd require a robust and simple means of using available aircraft data to do that. And then address various what-if scenarios based on failed or misleading data on any of those selected parameters.
You'd also have to ensure that you didn't convert a float into a hard landing, or indeed a normal landing into a hard landing. Spoiler deployment at low altitudes is recipe for trouble. So again, you'd need to have a system that didn't create new risks; which might mean limiting the authority to such an extent that it wouldn't be able to do much.
And philosophically, if we're going to say "the pilot won't land properly, let's add a gizmo to compensate" - it's simpler (and the technology does exist now) to just say "take the pilot out of the equation, engage autoland".