How does a short-term A/B test measure long-term negative sentiment? Measuring minutes played or cancellations over a few months is pretty easy. Measuring cont

How does a short-term A/B test measure long-term negative sentiment? Measuring minutes played or cancellations over a few months is pretty easy. Measuring contribution to a gradual loss of trust, fewer or weaker recommendations to others, and perhaps cancellations months or years later is all but impossible. It's really hard to construct a split test that's deep enough to fully measure negative reactions. Usually that's not an issue because the thing being tested either has a positive or neutral impact; other than by accident, it's not downright offensive. Fully measuring something that a meaningful percentage of users consider offensive would be really tough. It's even possible for a decrease in trust that doesn't manifest in user behavior for years. Maybe they stay customers and continue playing media, but start treating Netflix's future decisions with skepticism. The closest to an indicator is probably NPS, but NPS has its own issues. (The fact that Netflix isn't testing a user-facing "Show a preroll trailer from Netflix's best recommendation" setting says a lot about their willingness to ignore long-term, hard-to-measure impacts…)