Say that you have made a 30" TV commercial.
That TV commercial runs @ 25fps.
Clients also needs a version for Cinema.
DCP is running @ 24fps.
So the 25fps video will be running slower in cinema, and as a result will run for 31,250 seconds instead of 30". (-4%)
You, on the other hand, need to deliver audio that is 31.250 seconds long. (to match the longer video-running time)
There are different ways to do that.
The most common practice is stretching the audio. Making it longer, and pitching it up again.
Using pull, you can actually “pull” your audio session, so it runs for 31,250 seconds, thereby eliminating the stretch-process.
(You are actually faking that your session runs at slightly lower sampling frequency than 48kHz)
You will however need to (re)record the output of your session into another recorder.
Like it was done in the old times - re-recording the masters on another machine-
Exporting or re-recording within the same project will not work, since you are still locked to your 48kHz Houseclock.