I used to do talks like this for city planners. These were my main bullet points:
Prepare the map for printing instead of display. Think about line weight and type size when the map is used in an 8.5 x 11 report, not how it will look as a 30 x 48-inch plot at the meeting.
Start with a good base map. Less of a concern now, in the full GIS era, than it was back then. But people today use data layers—just because they have them—that are only of marginal importance to the point of the map. If you're mapping the cities where NACIS has been held, is shaded relief relevant?
Use data at an appropriate scale. You can't use shorelines digitized at 1:1 million on a map showing the city's waterfront at 1:4800. Similarly, those mountain roads digitized at 1:10,000 will pile up on each other at 1:250,000. Time to introduce generalization, aggregation, and displacement.
Don't be afraid to redraw. Still the thing I think would make the biggest difference in most maps. Once you've moved the map over to the Illustrator side, look at whether you could dramatically improve the look of the map by redrawing a few important but overly fussy lines: the city limits, an overly elaborate water feature or park, freeways or railroads that need to be smooth curves, long streets or boundaries that need to look straight.
Generalize. Related to the last point, but like shouting into the wind in the GIS era. But let me quote 1988 Dennis: "Every jot and tittle of a city's legal boundaries need not be shown, for instance, on a simple map locating a neighborhood within the city. In fact, it's distracting."
Use standard symbols. US highways have a distinctive shield, which is different from Interstates, and both are different from state shields. I was also railing back then against the inappropriate use of fill patterns that made no sense, such as Zipatone film with little tufts of grass to represent industrial districts.
Symbols may not be the best way to present data. Often it's better to just label every library than to invent some inscrutable symbol. If there's a symbol in the legend that only appears once on the map, time to rethink. And never force the reader to look at a numbered list if the labels will fit on the map.
Consider small multiples. Rather than cramming everything onto one big map, think about whether your point might be better made with a series of small multiples. Libraries and fire stations aren't interchangeable; why show them on the same map if you're looking for areas that lack one or the other? Even for land-use maps, perhaps the point is that most of the multifamily is east of the freeway, and a map for each type of use would better show the patterns than one big multicolored map.
Think negative as well as positive. We're so used to sketching with dark ink on white paper that we don't think about things like a gray or neutral-color ground, with some background information (like county lines or side streets) reversed out, and more important info in positive dark colors on top.
Think about the limits of reproduction. Back then I was talking about registration, the advantages of working in negative, and other quaint artifacts. But remember things like press gain, and beware of RGB color formulas that might get printed (unpredictably) in CMYK, or fonts that aren't incorporated in the final file sent to press.
Observe lettering conventions. Water features in italics, big features like mountain ranges letterspaced (but only if they're all caps).
Watch those patterns (and colors). Back then, people used patterns that didn't always have a clear progression. Today, they want to wander through the entire crayon box to differentiate thematic map districts, when they should be ranking those in a clear progression. Introduce them to ColorBrewer and sequential vs. qualitative. A grayed color palette can give a map sophistication that grocery-store reds and blues don't have.
Is a choropleth map appropriate? Or is the size of your districts overwhelming the differences in the data? The old familiar red-blue election night map problem. Alternatives: dot-density, bar charts or graduated circles on the districts.
Choose appropriate data class breaks. Quintiles vs. natural breaks vs. equal classes, etc.
Finally, my three pet peeves about marginalia: North-arrow worship (more attention apparently paid to that than the map itself); legends headed Legend; and class breaks like 200.0-499.9 that imply a false precision.