I've been struggling with this same question the past two years by scaling up a machine learning team (our job was to design and put into production various machine learning-based products for the company). When the team was small, there was no way we could create a product team (no enough product managers or machine learning engineers). So we started with skills. But it came with its lot of inconvenience: hard to align people across the various roadmap. Some teams were swamped and had a hard time prioritizing. Beside, teams by skills tend not to understand their constraints and way of working (thing product manager vs developer culture clash). Honestly, I feel that the product organization is the more efficient one in delivery and team alignment. But this requires vast resources if you manage a lot of different products (this was our case). And in the end, you can end up with doing some things twice or not investing enough in the common assets/architecture (my team was in charge of managing our data platform as well, so we had to think long-term and global at the same time). I quit my job by if I were still here, I would definitively cut back on the number of projects and rearrange the team by products.