Say we value article views and user signups. If I'm taking actions that achieve n views for each lost signup and you're taking actions that achieve m signups per lost view, where m < 1/n, then we could get a pareto improvement by doing less of both of these things.