I think for a good share of products it isn't necessary, but I also guess that it's very hard to see from the start. I strongly believe there generally is value in AI, but finding this particular application that unlocks it -- is a conundrum. So, if I imagine myself being a head/vp of product, in one of these companies, I have pretty much no choice but to integrate it and study the metrics. Then, after a few iterations the value is either found or not, and depending on the sanity of stakeholders, the useless AI features can be retired. But not before the evidence is collected, because you can't fight the AI frenzy without ones, lol