Theory and early simulation results suggested that the implementation of genomic selection should result in a lower rate of inbreeding per generation. However, current experience has shown that sires selected on GEBV have a higher inbreeding coefficient than those selected using conventional approaches. With a large proportion of dairy populations genotyped genomic information offers a unique opportunity to understand and manage genomic diversity. With this work, we will show different approaches aimed at constraining inbreeding accumulation while accounting for the heterogeneity of inbreeding depression across the genome. In most cases characterizing and managing diversity and inbreeding using autozygous segments, either homozygous by descent (HBD) or approximated by runs of homozygosity (ROH) is advantageous compared with using average genomic inbreeding metrics. In simulated data constraining the accumulation of long HBD (ROH) allow more efficient recruitment of new potential beneficial haplotypes as well as limiting the accumulation of deleterious variants in the population more efficiently. Identification of specific variants related to recessive load remains difficult. Using a heuristic approach, an average of 244 deleterious haplotypes are flagged as potentially partially detrimental per each combination of chromosome-trait of interest. Many of these haplotypes are shared across traits and have a small overall effect, in accordance with the partial recessive theory of inbreeding depression.