23&Me convinced me this process is mostly garbage. If at most you can conclude today that a complex phenotype has a complex underlying genotype: you're just confirming what we knew two decades ago, and not providing any useful answers for how the complex phenotypes come about, just that variations of the genotype lead to somewhat different outcomes. Totally boring.
Even if something is actually "genetically" heritable, we don't how heritable it would be over the space of all possible environments, just the ones we test.
Yes, an association to a phenotype using data like that sourced from 23&me only generally points in the direction of the genes and snvs which cause alterations in phenotype, but that is just a stepping stone.
It narrows down the search space to about 1-2% of the genome, which then follow up studies can fine map. Using high resolution data, such as whole-genome or exome sequence, you can then pinpoint protein-coding changes which alter phenotype. This is what a lot of people are doing right now, using the last decade of GWAS results to fine map using WGS.
If you discover exactly how the machine breaks, it is a lot easier to fix it, or at least prevent it from getting broken in the first place.
It is not a simple as flipping one switch, and it is an incremental 'boring' process, but the potential to better understand disease today and in the future is worth its weight in gold.
Obviously this isn't true for some diseases, such as simple mendelian diseases caused by single SNP changes.
I worked in the field for decades and very little progress has been made mapping genotype to complex organismal phenotypes or appreciating how complex development processes are affected by collections of mutations. This specific research, even if wildly successful, wouldn't really have impact on any significant diseases.