How can Erlang C values be greater than one? Here is an Erlang C plot I did myself. The x-axis represents the total traffic while the y-axis represents the probability. Each line represents a different number of total available channels.

As you can see, the graph has probabilities greater than one. How can this be even possible?
 A: The formula is only valid under the assumption $A<c$.
In the formula, $A$ measures the average intensity (e.g., in the rather dated call-center application, the average number of calls you have at any given moment), while $c$ measures the number of channels (in the same model, it's the number of calls you can process at any given moment).
When $A<c$, the system will be ergodic: sometimes you will have calls (or whatever) pile up, but on average, they will be processed faster than they appear, and long-term behavior is stable. We can compute properties of the typical state of the system; in particular, the probability that all channels are busy.
When $A\ge c$, the expected long-term behavior of the system is divergent: calls keep piling up faster than they can be handled, so that the number of waiting calls is proportional to the time. This is not surprising: if $A = 20$ and $c=10$, then this says that you're getting an average of $20$ call-hours per hour, but can only process at most $10$ call-hours per hour, so you get $10$ extra call-hours per hour that went unhandled.
In this case, there is no "typical state" and the usual analysis we can do is nonsense. We should ignore the formula, which gives a probability greater than $1$ for $A>c$, and just say "our analysis of the system says that everything goes horribly wrong in this case". 
