Designing Effective Data Metrics: Definition, Elements, Evaluation, and System Construction

The purpose of metrics is to "learn more"—by assigning numeric values we can convey sufficient information at an acceptable cost.

Middle‑aged Jia visits a doctor who only says "okay" without giving blood pressure, body fat, or glucose numbers.

Drunk‑driving suspect Xiao Yi is asked how much he drank, but the officer lacks a blood‑alcohol metric to decide the penalty.

CEO asks for sales performance, and the VP replies "great" without any sales amount, per‑capita output, or trend data.

Without metrics, the information we obtain is extremely limited and costly to acquire.

01 What Is a Metric?

A metric is a defined numeric value used to quantify and abstract a fact. Simple facts (e.g., an object's weight) can be measured with a single metric, while more complex facts (e.g., a person's health) may require multiple metrics such as BMI or body‑fat rate. For highly complex facts like a country's economic condition, indicators like GDP involve many layers of abstraction and massive data, often requiring an entire metric system.

In summary, a good metric should contain at least four elements:

Name : clear and unambiguous.

Owner : the person responsible for maintaining and operating the metric.

Meaning : description of the quantified fact, its scenario, purpose, and what it characterizes.

Scope : definition of how the metric is calculated, data sources, and timeliness.

02 What Makes a Good Metric?

Metrics can be evaluated on four dimensions:

Effectiveness : Does the metric reflect the fact we want to quantify? (e.g., choosing DAU vs. MAU for different app types).

Credibility : Is the metric stable for the same fact? (e.g., consistent test scores for the same candidate).

Sensitivity : Can the metric capture changes in the fact promptly? (e.g., using "no‑room‑available rate" for hotel bookings).

Operability : Can the metric be used in daily operations to drive improvement? (e.g., interpreting a 10% drop in NPS).

03 How to Design and Apply a Data Metric System?

Building and operationalizing a metric system raises questions such as how to construct the system, how to identify bottlenecks through data analysis, and what actions to take afterward.

To help readers systematically learn the concepts, methods, and practical applications of data metric systems, DataFun released an ebook titled "Data Metric System: From Design to Implementation" . The ebook is divided into two parts: (1) methods for building metric systems, covering technical points, challenges, and solutions; (2) industry‑specific applications, featuring cases from Douyin, Kuaishou, Didi, JD, and finance, as well as advanced topics like intelligent metric platforms and automation.

Ebook Highlights

Method + application: theory demonstrated through practice.

Rich industry cases from leading companies.

Insights from senior experts and chief technical advisors.

Table of Contents (selected)

How to build a good data metric system?

Kuaishou metric system practice

Douyin group metric analysis and growth practice

Didi metric system construction practice

Metric and label applications in finance

JD business metric system practice

Intelligent metric platform construction and automation

Kuaishou metric standardization and OneService platformization

The ebook can be obtained by scanning the QR code and sending the keyword "指标" to the official account.

Authors

Xu Yao – Chief Technical Advisor, Zhiguang Technology

Qian Yingnan – Data Product Director, Kuaishou

Bao Wenxia – Senior Data Analyst, Volcano Engine

Cao Lei – Senior Expert Engineer, Didi

Zhang Kun – Product Director, Kyligence Open Platform

Zhang Wanqi – Data Mining Engineer, JD

Zong Zheng – Senior Technical Evangelist, Kyligence

Liu Yifan – Head of Kuaishou Data Service Toolchain Team

Reference: "Metric System: Design Methods!" (https://mp.weixin.qq.com/s/TioHEFb4mWHY6KItbGxPXw).