Inference based on a single observation Imagine we use machine A to perform a task, we repeat it 1000 times and it always takes more then 30 min to finish the task.
We buy a new machine (machine B) and in the first run it takes 29 min to finish the exact same task. Can we claim machine B is faster than machine A?
What if we make the numbers more extreme: machine A was tested 1,000,000 times and it always took more than 30 min to finish; it took machine B only 20 min to finish the same task.
 A: It boils down to what you're prepared to assume for the second case. 
For example, if you assume that the shape of the distribution would remain identical but the distribution could be scaled by an unknown multiplicative constant, then you'd be able conclude that this multiplicative constant would be smaller than 1. 
(Similarly if you assumed that the only change could be a shift)
However, if the two distributions might be very different in shape - or there might be both a shift and a rescaling - then you'd have little basis to rule out the situation whuber raised (the next observation from B may well be far above all the other observations).
With only one data point your conclusions are not at all robust to such assumptions as might allow you to conclude there's a reduction, but we're not in any position to tell you what is reasonable for you to assume (we don't know the circumstances).
A: Depends on the SD (distribution) of machine B (which is unknown at this point), extreme case:

So we cannot claim that B is faster.
