Understanding Data Product Layers: Business Value, Data, Algorithms, and Applications

The application layer of a data product focuses on delivering value to its target audience; a product’s quality is judged by how well it creates real, measurable business impact.

Using the "5G infrared imaging temperature measurement" system as an example, the article describes how the product is deployed in public venues (airports, train stations) to non‑intrusively and quickly identify high‑temperature individuals, a critical capability during the 2020 COVID‑19 pandemic. In the data layer, large, high‑accuracy training datasets are collected by pairing infrared thermal images with temperature readings. In the algorithm layer, face‑recognition and infrared imaging techniques, often powered by neural networks, are trained to predict body temperature, with distance between subject and sensor significantly affecting accuracy.

The discussion then presents two scenarios. First, data with strong certainty for business value, exemplified by the ImageNet dataset, which, despite its complexity, offers low uncertainty and can drive commercial products when sufficient compute and algorithmic advances are available. Second, data with high uncertainty, where it is unclear which data are truly important or whether they can be collected—illustrated by challenges in medical diagnosis, ambiguous disease definitions, and privacy regulations that limit data usage.

The article highlights three methodological pillars for strategy in domains such as ride‑hailing: experimental design and causal inference, and reinforcement learning. Experimental design enables the collection of useful data for causal analysis, allowing businesses to identify and manipulate key variables to achieve desired outcomes. Reinforcement learning seeks optimal long‑term reward policies, leveraging rich user trajectory data (order history, call records, coupon usage) to predict the long‑term payoff of different actions for drivers and passengers.

The algorithm layer serves as the bridge between data and application, requiring deeper features and advanced techniques—such as causal inference and reinforcement learning—for high‑impact decisions. Moreover, sophisticated algorithms can compensate for gaps in data collection by mining existing data to uncover insights and guide future data‑building efforts.

Special thanks are extended to contributors for their valuable insights.