Just look at random "not so well formed" Word file like for example:
https://arxiv.org/pdf/1902.07133.pdf
Not commenting the bad typography of text, math, tables, etc., just try to estimate the effort for converting the references in well-formed and validated BibTeX entries.
I know that a lot of smart scientists struggle with computers and they cling on MS word because they have invested a lot of time and effort into learning how to use it. But I think, we should try to convince them that there are better, simpler and more reliable ways to author papers.
https://ac.els-cdn.com/S1303070118300945/1-s2.0-S13030701183...
The 3B2 PDF looks much better than a Word file and especially the references are all clean, linked, and downloadable as a BibTeX file. There is also a HTML version of the work. I can see definitely a lot of added value to the scientific document there.
You fail to see that nonprofit publishers (IEEE, AIP, APS, AMS, etc.) also typeset and copyedit articles just like Elsevier, and that too at a much smaller price. The point is that you don't have to charge customers exorbitantly if the only thing they are gaining on publishing with Elsevier is neatly typeset articles. Further, Elsevier outsources the production of its papers to companies [1] in India. If it's production quality that one wants, those companies can probably do it at a fiftieth of the price Elsevier finally charges the customer.