Instead of automating the checks, minimize them by reducing the guidelines!
The critical pre-submission guidelines are only those where papers that don't follow them would have to be severely edited in a way that will change the referee's opinion about the paper. This is, e.g. a draft that is twice the permitted length, or one where the section organization used is not permitted. Many newer journals are pretty good about things like this (eLife, for instance: https://elifesciences.org/inside-elife/a43f95ca/elife-references-yes-we-take-any-format-no-we-re-not-rekeying)
If I took guidelines 100% seriously, I would be spending a great deal of time not just on margins and formatting (mostly fast once I have a template), but on things like:
- Does this journal use Fig., fig., or Figure?
- Should subfigure captions be a) or A) or a or A or a or A
- Should journal names be abbreviated?
- Are the acknowledgments I wrote allowed?
- Are footnotes permitted, or is everything an endnote?
- Do figure captions have titles?
- Is it appropriate to use a comma in an axis label (1000 vs 1,000)?
- Is it Eq. 1, Eqn. 1, Eq. (1), ...
Eventually these have to be settled by work from the authors and the editors - but I've found that even different copy editors at the same journal can have different interpretations!
I view rejecting a submission for issues like this as being a waste of everyone's time. With an automated system, it might be less of a waste of time - but I think it also encourages journals to be nitpicky at a stage when it is not important.