Love this paper! The effect of autoenrollment isn't static, it erodes over time due to leakage, job transitions, etc. A lifecycle model matching this attenuation job transition effect predicts 3% AE in all jobs ↑ retirement consumption by just ~0.3% (though 2x for low-earners)
The most famous "nudge" example -- defaulting to 401k enrollment with auto-escalation -- doesn't actually do very much it turns out.