HLS streaming and bandwidth efficiency

Question

I am observing when playing a HLS playlist that the player will go ahead and download the entire content as soon as it can rather then get segments just in time or perhaps only 2 or 3 ahead.

This is a problem because it is inefficient in terms of the volume of data pulled from the CDN. If a user plays only 10% of a piece of content but downloads 100% of the data CDN costs will be much higher than they might be if the player only loaded a modest amount of data ahead of time.

What techniques are available to introduce a more conservative approach to data download? I'm more interested in mobile but the question applies equality to web.