Automated vectorization of legacy maps has been a 'Holy Grail' since the origins of GIS in 1960s ( https://www.cia.gov/stories/story/the-mapmakers-craft-a-history-of-cartography-at-cia/#the-1960s ) - and sixty years later, it is still a persistent problem. And it still has the same parameters, the ROI of automation vs. manual, i.e. are you doing 10 or 10,000? What is the quality of the scanning? How dense and consistent is the symbology? How similar are they to one another? For instance a utility company's archive produced by the same drafting department with standard rules might be very tractable for automation, while 10 medieval manuscripts would be a career by them selves.

There are existing opensource and commercial products already, so why not test those first to see if they meet your quality threshold - for instance just gathering approximate areas versus a high quality cartographic product. There are tools outside of GIS also, I've used Adobe Photoshop to very good result because it has baked in preview tools for tuning transformations, despeckling, etc. - and dealing with half-tones and hachures is somewhat possible ( https://en.wikipedia.org/wiki/Hachure_map ).

Automation introduced a large unrecognized unpredictable QA task and tuning iterations, while manual digitizing can be linearly estimated by doing a small sample and extrapolating. So don't forget https://www.mturk.com/