Bet this isn't a PMax-versus-Shopping problem at all. More likely you reset the learning phase and sliced your conversion data way too thin in one go.
Feed-only PMax is getting sunset anyway, so the old setup was doomed regardless. the real pain hits when those segmented Shopping campaigns go live on max conversion value bidding, because every campaign now has to relearn from a whisper-thin sliver of data. That kind of data starvation almost always produces a scary-looking short-term dip.
here's the playbook I'd run:
Consolidate first. One campaign for best sellers, one for everything else, not a segment per product. each campaign needs enough conversions to claw its way out of learning.
Wait it out. Give it two or three conversion cycles before judging anything, and pull up the conversion lag report while you're at it. some of that "drop" could be attribution lag playing tricks on you.
Compare honestly. Like-for-like date ranges only, and double-check nothing else shifted: budgets, feed quality, seasonality.
Segmentation looks gorgeous when the volume is there to feed it. until then, consolidation usually wins, and it isn't even close.
