Well, anything that if you do incorrectly someone (or some robot) will likely spot it before the failure gets to a customer. For instance, if you have a robot that picks fruit but 1% of the time the fruit is not ripe or is worm-eaten or something, then you can have a process where the picked fruit are inspected a second time and the bad ones thrown out.
Another case where 1% failure would be acceptable is weeding. Suppose you have a bunch of robots removing invasive English ivy from a forest. Maybe once in a while they'll pull up salal or a fern or something by mistake, but that's fine because those aren't endangered species and in the absence of ivy the ecosystem will quickly recover.
I worked in a factory that was the distribution centre for all the cold and frozen food in every Safeway in the sate. There were a few hundred of us who had to pickup a heavy box from a conveyor belt, turn around, take a step or two and put the heavy box down on a pallet. When the pallet was full someone would come and take it away. Repeat. For 8-10 hours a day, 6 days a week.
Upstairs people were manually putting those boxes on the belts. It destroyed people's bodies, I'm happy I only worked there for a few months each year to put myself through University.
The factory was very high tech and cutting edge, though the humans at each end of the conveyor belts were unavoidable.
We messed up all the time - dropped boxes, did bad stacking and made pallets fall over, crashed pallet jacks and knocked over pallets etc. The failure rate was probably around 5%, I'd guess.
Starting pay was around $30/hr, double that on Sundays, and 2.5 on Public Holidays
A humanoid robot that actually works most of the time will change the world.
This doesn't seem like the kind of problem that would require a humanoid robot to solve though? It's just that even with current state of the art robotics, things like grabbing cardboard boxes (easily crumpled or dropped, mass distribution within may be uneven or unstable, may already be broken or leaking etc) and stacking them into a stable 3D pallet is actually quite hard, so much so that Boston Dynamics managing to do some parts of this in optimal conditions with uniform, lightweight boxes was hailed as a breakthrough:
For many menial tasks, a 1% failure rate (or indeed any failure rate higher than a human) is generally not a problem so long as the economics stack up and safety isn't a concern.
Every manual job I’ve ever done I’ve probably failed more than 1% of every movement. But then I correct. In a car moving at high speeds (focus matters more), I do this less often, but I have been in crashes.
Another case where 1% failure would be acceptable is weeding. Suppose you have a bunch of robots removing invasive English ivy from a forest. Maybe once in a while they'll pull up salal or a fern or something by mistake, but that's fine because those aren't endangered species and in the absence of ivy the ecosystem will quickly recover.