Cold-Start Weeks That Actually Teach the Model

Cold-start policies often hide behind popularity fallbacks. That keeps empty slates rare, but it also delays the moment the system learns anything personal about the user or the item.

Measure how long it takes for a new account to receive a majority-personalized slate, and how long a new item waits before it appears outside promotional rows. Those two clocks explain more product frustration than aggregate precision charts.

Onboarding prompts should collect preferences the ranking layer can actually consume. Taste quizzes that never map to features are theatre; short, mapped signals compound.

When we study cold-start weeks with UK product teams, the breakthrough is usually a clearer first-week scorecard—not a larger model.