Note that the metric for "economic success of FSD" is even more favorable than the '99%' metric: FSD has to work much better than humans,
on a significant percentage of routes that matter economically to a specific customer.
For example FSD working well on highways is already a big step forward, because it already covers around 50% of Vehicle Miles Traveled (VMT):
Note how big proportion of vehicle miles traveled highways are, compared to 'local' roads, yet they are just a small part of the road network.
Also note that highways are also the most dangerous ones in terms of traffic accident fatalities (due to the much higher speeds and quadratic kinetic energy), so Tesla starting with NoA also improved safety disproportionately.
Likewise, if for example a trucking company's fixed routes of travel are handled well by FSD then it's only those ~1,000 miles of routes that matter to them, out of trillions of miles of possible routes. If FSD works on those routes then then FSD will be 100% functional and 100% economically useful to
them.
Tesla will also be able to concentrate their efforts on the AP disengagement events that a larger percentage of the Tesla fleet is reporting, so instead of going for a 99.999999% solution they can concentrate the effort for where it matters most.
So FSD is a statistical problem on a
per customer basis, with an automated feedback channel of disengagement events to Tesla, which feedback loop I expect to drive FSD market penetration a lot faster than many would expect.