Designing and Implementing a Unified Data Metric System for 58 Commercial Data Team

Background Digital operations have become a key driver for enterprises, but building a data middle platform that satisfies product, development, and operations teams faces challenges such as inconsistent metric definitions and unclear data sources.

Data metrics are metadata that quantify business units, serving as the bridge between business goals and data analysis. Since 2019, the commercial data team has standardized all existing metrics, created a metric white‑paper, and resolved inconsistencies.

Common Problems in Metric Definition

Same metric name, different definitions (e.g., multiple definitions of "purchasing users").

Same definition, different names (e.g., "sex" vs. "gender").

Unclear or incorrect metric descriptions (e.g., ambiguous "cash+discount" for "spending amount").

Inconsistent naming that hides business logic (e.g., CVR calculated differently across products).

Unclear data source or calculation logic, especially when front‑end and back‑end collection differ.

How to Standardize Metric Definitions

1. Domain‑driven Management : Organize metrics by business line, domain, and process (e.g., recruitment, real‑estate, automotive). Each domain aligns with data‑warehouse concepts such as revenue, traffic, and customers.

2. Atomic vs. Derived Metrics : Atomic metrics cannot be further split; derived metrics are built from atomic ones using period, granularity, and business constraints.

3. Naming Conventions : Use "action + measure" for atomic metrics (e.g., "registered users"); for derived metrics, prepend modifiers (time period, granularity) before the atomic name (e.g., "30d_product_purchase_users").

Metric Management System

The system, built on a metadata center, synchronizes domain and process definitions, creates metrics with standardized tags, and links them to source tables for easy discovery via a data map.

Metrics are stored with details such as English/Chinese names, calculation logic (e.g., SUM(CASH) ), source tables, and formatting rules.

Data Application Architecture

The platform consists of four layers:

Query Engine Layer : Supports multiple JDBC sources (Kylin, Doris) and connection pooling.

Metric Layer : Provides a generic SQL engine for metric calculations, supporting custom implementations.

Cache Layer : Caches query results to improve performance.

Report Layer : Offers visual components and ad‑hoc analysis.

Technical Implementation of the Metric Layer

Interfaces define common calculations (e.g., day‑over‑day, week‑over‑week). Base classes implement generic logic, while custom metrics extend the base class and override methods.

// Example of a custom metric implementation in Java
public class Imp extends BaseAbstractTargetService {
    public Imp(final Long id, final ViewExecuteParam executeParam, final User user) {
        super(id, executeParam, user);
    }
    /**
     * Override to compute day‑over‑day ratio.
     */
    @Override
    public String getRiHuanBi(TargetDic targetDic) throws SQLException {
        // Parameter extraction and SQL assembly omitted for brevity
        return StrUtil.format(excuteSql, paramMap);
    }
}

Metric Query API

The platform exposes a unified API where callers specify category, time range, dimensions, filters, aggregations, and sorting. The API returns column metadata and result rows in JSON.

{
  "cate1":1, // primary category (e.g., revenue)
  "cate2":2, // secondary (e.g., daily)
  "start":"20210401",
  "end":"20210430",
  "groups":["product_id","platform"],
  "filters":[{"name":"platform","operator":"in","value":["APP","WEB"]}],
  "aggregators":[{"column":"pv0","func":"sum"}]
}

Application Cases

Using the unified metric system, the team built a recruitment analysis dashboard that defines metrics, configures queries, and visualizes results through the report layer.

Conclusion

By combining systematic governance with a flexible technical stack, the team achieved consistent metric definitions, a reusable metric management layer, a unified API for diverse analytical needs, and rich visualizations that accelerate business decision‑making.