I made ZIP boundaries from the TIGER database in 1993. (Giant Delaunay triangulation/ Voronoi diagram. Fun algorithms for dealing with millions of points! Then merge polygons.)

I believe the demand for this was to visualize attributes on a map by coloring polygons, and the dataset was addresses that already had zip codes.

(Please stop coloring polygons by attribute: it gives false impressions because areas are unequal. Instead, use a scale of dot density. Or marker size, which has its own perceptual issues.)

However misguided, coloring zip code polygons gives you things like maps of customer data. But you could just as well geocode each address and assign it to a county or other boundary. (news flash: the geocoding database is derived from the same line data as the ZIP code project.)