Daily active users (DAU) counts distinct users who engage with a product during a given day. For a short-video app, it serves as a high-level measure of user return and product traction. section 7.1.1
DAU can guide product optimization, but it should not become the sole target. A ranking model may raise DAU while its underlying prediction metric improves only marginally, so business impact and model quality must be evaluated separately. section 7.1.1
Short-video ranking
An app initially shows the newest video first. Engineers predict the probability that a user will swipe away, then rank videos from lowest to highest predicted swipe probability. An A/B test shows higher DAU for the personalized version, making DAU useful for evaluating the product change. section 7.1.1
Figure 7.1 As the quality of the predictions, P{swipe}, produced by your ML model increases, the DAU increases, too, but at an ever-slower rate.
Use the figure to inspect how improving prediction quality can increase DAU at an ever-slower rate. section 7.1.1
Interpret DAU alongside metrics that represent the business’s specific goals and possible harms. An advertising system may increase short-term revenue by showing more ads while reducing DAU or time spent per user over the long term. section 7.2.1 section 7.3.1
