Basically that means looking at voters' value systems (why they do what they do). And you find factors that are the best predictors of voters' behaviors. Segmentation would cut voters into as few distinct groups as possible. Guessing (because I don't have the data):
27% hyper-partisan Dems, people who will turn out in every election, every primary and vote Dem. No persuasion or pandering needed.
17% fair weather Dems -- vote in most elections and some primaries. Not watching the play-by-play. Persuasion needed.
10% old school Dems -- want the party to support unions, wages and so soc. FDR-style policies
9% identity politics Dem -- view the Dems as better on race and gender issues. Top priority is representation for their identity
8% Reagan / Cheney / Lincoln Project Dems -- could vote GOP but could vote Dem if attracted by GOP converts
etc.
Then you would rank candidates by how many of those segments they can move and give preference to those who get the highest numeric totals. eg candidates who could win in a GE
Then you would hold primaries among those who made the cut and have them to compete in speaking skills and crowd skills.