In my experience, attempting to improve code coverage of legacy software is usually premature.
That's because it's often definitely known that the feature coverage is incomplete: there are untested features.
In that circumstance, measuring code coverage may be harmless, and can be helpful. (Hint: the coverage should be improving.) But the better guide at that point is the feature coverage.
So: pay primary attention to feature coverage, secondary attention to code coverage.
And always (as my boss DonD used to say), "Run your business not by the numbers, but with attention to the numbers."