How to Build an Effective Indicator System: From Concept to Productization

01 Indicator System Problem Definition

Metrics are statistical concepts that reflect overall quantitative characteristics; in a data warehouse they are core artifacts that must be accurate, complete, timely, and consistent. An indicator system is an organic whole composed of interdependent components, not an isolated technology.

The system has three layers:

Concept layer – carries core business concepts such as transaction volume or daily active users.

Process/mechanism layer – ensures the authority and continuity of concepts through defined workflows.

Productization layer – typically a content product (e.g., an indicator platform) that implements the concepts.

02 Indicator System Design

1. Achieving Concept Consensus

Concept consensus is the foundation; without it the system is an empty tower. It can be driven by BI bridging business and data engineers, or by data architects who have a holistic view. When full‑member consensus is impossible, consensus is achieved within roles and translated by engineers.

2. Position of the Semantic Layer

Three approaches exist:

Embedding the semantic layer in the data warehouse – low upfront cost but limited agility and heavy coupling.

Separating the semantic layer as an independent product – decouples logic from physical storage, supports multiple lake‑warehouse scenarios, but incurs higher initial cost.

Integrating the semantic layer into consumption tools – highly flexible but may lack consistency across platforms.

3. Building Concept Consensus

OneData methodology standardizes metric language, but it lacks macro‑level perspective. Introducing Domain‑Driven Design adds a domain model and ubiquitous language, helping to partition data domains and align business terminology.

03 Mechanism & Process Design

Mechanisms ensure continuous construction and freshness of metrics. Challenges include reliance on offline Excel maintenance and limited consumer usage, which break the positive feedback loop. Two solutions are proposed: productizing and applying AI to improve management efficiency, and deep integration with consumption platforms to speed up data retrieval.

Roles and responsibilities:

Business owner – defines requirements, owns metric definitions, ensures unambiguous standards.

Technical owner – implements metric models, maintains alignment with business owners.

Consumer – final user who must be informed of metric changes and use data responsibly.

A change‑process is designed where business owners have final approval and consumers receive notifications.

04 Indicator System Design – Productization

The productization carrier is an indicator platform that addresses efficiency for data managers, modelers, developers, and consumers.

Standardized metric definition and management to ensure accuracy and consistency.

Efficient metric development to quickly respond to business needs.

Convenient consumption capabilities for easy data access.

Drill‑down analysis to locate issues and optimize.

Typical platform structure includes unified terminology management, atomic metric management, and business‑limited management, enabling derivation of a large metric library and services such as data query APIs.

Current shortcomings:

Only logical metric definitions are standardized; physical definitions hidden in complex SQL, limiting multi‑metric ambiguity resolution.

Drill‑down analysis is difficult due to complex SQL in ADM/DWS.

R&D efficiency needs improvement for complex scenarios.

General consumption is limited to metric‑code APIs, lacking table‑style access.

05 Business Practice and Future Outlook

Practices have produced over 30,000 derived metrics, with >70% automated. Automation and codeless definitions have reduced ambiguity and improved performance, achieving >10× R&D efficiency and ~20% cost reduction.

In specific domains (e.g., Netbank), challenges include metric‑definition consistency, agility, and analysis difficulty. Unified data models and deep integration with analysis platforms have enabled thousands of derived metrics with >90% automation, shortening delivery from days to hours and cutting costs by ~30%.

Future focus will be on leveraging large models for the semantic layer, assisting modeling, reducing construction cost, and enabling natural‑language data discovery and analysis.