Understanding the Three Core Layers of a Data Product: Application, Data, and Algorithm

A successful data product consists of three core layers: an application layer (the business goal), a data layer, and an algorithm layer.

Example: the "5G infrared imaging temperature measurement" product is deployed in public places such as airports and train stations to identify high‑temperature individuals without contact, a critical business goal for COVID‑19 prevention in 2020. To achieve this, the data layer collects large, high‑accuracy training datasets of infrared images and temperature readings, while the algorithm layer uses facial‑recognition techniques and neural‑network models to predict body temperature, noting that the distance between a person and the imaging device significantly affects accuracy.

Three core layers :

Application layer: Implements technology to meet the business objective, guiding the data and algorithm layers.

Data layer: Builds efficient, orderly low‑level data foundations driven by business needs, facilitating extraction, cleaning, and reducing analysis difficulty.

Algorithm layer: Provides theoretical research and techniques (sampling, experimental design, MCMC, linear models, random forest, SVM, neural networks, deep learning, causal inference, reinforcement learning) to achieve the business goal and bridge data and application.

Various examples illustrate how value is created: developing R packages for statistical analysis, medical research influencing treatment decisions, kidney‑transplant matching improving success rates, government reports shaping policy, e‑commerce platforms linking supply and demand, and education platforms offering high‑quality resources.

Key considerations for the data layer include ensuring data serves the application, balancing collection cost against benefit, and transforming data into actionable business strategies.

The algorithm layer discussion covers regression models, random forest/XGBoost, and deep learning, highlighting their success in image, speech, and natural‑language tasks. It also emphasizes causal inference and reinforcement learning for platform optimization, such as ride‑hailing services that use historical trajectories to design pricing and dispatch strategies.

Conclusion : Data products can be classified as survival‑type, service‑type, or quality‑type. A high‑level data product is guided by the application layer, builds an economical data framework, and integrates the three layers to deliver valuable solutions to users, enterprises, or governments.