Machine learning (ML) and artificial intelligence (AI) have rapidly expanded in real-world applications, but their use in product analytics can be problematic, as exemplified by Mixpanel's experience. While ML features can demonstrate impressive results in demos, they often lack transparency and produce confusing or misleading outcomes, leading Mixpanel to de-emphasize their development. Predictive models, commonly used to influence user outcomes like retention, often misinterpret correlations as causations and fail to provide actionable insights. Similarly, clustering algorithms used for user segmentation can be arbitrary, lacking context and producing unreliable clusters. To address these challenges, Mixpanel suggests alternatives such as focusing on correlation rather than prediction for understanding causal drivers, and employing qualitative research or pragmatic segmentation methods for user categorization, highlighting the limitations of ML features in analytics products.