Early in my tenure as PM for our mobile checkout flow, I pushed to add a guest checkout option based on user research showing friction at the login step. We shipped it, and conversion did improve slightly — but cart abandonment on the back end actually increased because guest users didn't save payment info and dropped off at confirmation. I should have looked at the full funnel, not just the login step. We eventually rolled it out with a smarter save-card prompt and recovered the numbers. It was a good learning experience about thinking end-to-end.
I made the call to de-prioritize offline mode for our mobile app, arguing that less than eight percent of sessions showed connectivity issues in our telemetry. Three months post-launch, NPS for users in low-bandwidth regions dropped eleven points — a segment our aggregate data had masked. My error was trusting average connectivity metrics instead of segmenting by geography before deciding. I owned a post-mortem, identified the segment gap, and introduced a standing rule on my team: any decision affecting infrastructure must include a segmented usage cut before sign-off. That rule is still in the team's decision checklist today.