Well, it simply means that based on the sort of chances we allow/concede (the statistical probability of those chances resulting in actual goals against), we concede more goals than we "should" (statistically and based on the metrics they use).
Similarly, if you under-perform in the xG (expected goals) department, you create an amount of chances which "should" result in more goals than you actually score.
Whether you buy into it (fully or partially) is a different matter. But the idea is that a team that under-performs per xG/xGA will sooner or later revert to a mean of sorts - just as a team that over-performs in that regard will do the same. However, xG as a prediction tool in isolated seasons has proven to be quite unreliable in several cases.
An example: a team that actually defends well on the whole might concede a statistically improbable amount of goals - over a shorter period - from, say, low percentage (statistically) long range efforts or silly mistakes from players who aren't part of the defence as such. Such a team will likely revert to the mean over time (statistical anomalies don't last unless you're extremely unlucky).
Another example: our xGA in Jose's second season indicated that we didn't defend particularly well. Why did we over-perform? Because DDG pulled a number of statistically improbable saves over the course of the season.
Again, though - to conclude that results will improve (or the opposite) within the context of a singe season just because the xG looks good (or bad) would be, well, silly.
