This is an idea from a 'how do I' question from Erik Christoffersen which hadn't occurred to me before.
What he wanted to do was along the lines of having an underscore looping under a speech, and then on the last word of the speech de-vamp to a loud cue, but also build the level of the underscore, in the time remaining before the cue after the de-vamp triggered to match the level of the incoming loud cue.
This has the advantage of it appearing to the audience that something happens bang on cue after the last underscored word is spoken. Without this there is potentially a long wait while nothing happens.
Here's the solution I came up with.