KwaiBI: Evolution of Kuaishou’s One‑Stop Business Intelligence Platform from 1.0 to 2.0

01 KwaiBI Platform Introduction

This chapter provides a macro overview of Kuaishou’s one‑stop analysis platform, including its product capabilities and development history.

1.1 What is BI?

BI (Business Intelligence) refers to a complete technical solution that integrates data, enables data exploration, insight, analysis, reporting, and prediction, and drives business actions or strategy decisions.

1.2 KwaiBI – One‑Stop Self‑Service Data Analysis Platform

KwaiBI is Kuaishou’s internal name for its one‑stop BI platform, aiming to provide rich analysis tools, improve data analysis and decision‑making efficiency.

Current statistics: MAU ~15,000+, millions of reports and data models, and support for over 150 business applications.

1.3 KwaiBI Usage Scenarios

Kuaishou’s analysis system consists of two major platforms: the BI platform and the indicator middle‑platform. Their tight integration enables the indicator platform’s capabilities to be applied across BI tools.

The five core consumption scenarios are data acquisition, analysis, visualization, push, and portal creation, with flexible open‑integration capabilities.

1.4 KwaiBI Core Process

The core workflow is divided into four steps:

Data Ingestion: Users import data tables into the BI platform.

Data Preparation: Extract and process tables, drag‑and‑drop to build data models and datasets.

Data Analysis: Use datasets in analysis tools for interactive or code‑based analysis.

Data Consumption: Publish results as dashboards, screens, pushes, etc., for business consumption.

1.5 KwaiBI Product Preview

KwaiBI’s product matrix covers most analysis scenarios, including visual modeling, data acquisition tools, self‑service analysis tools, data push & smart alerts, and portal creation.

1.6 KwaiBI Evolution

KwaiBI has undergone four major stages:

Tool‑based (pre‑2019): Simple reporting tools could not keep up with rapid business growth.

Platform‑based V1.0 (2020): Built a one‑stop platform addressing five core consumption needs.

Standardized V2.0 (late 2020): Adopted headless‑BI concepts and deep integration with the indicator middle‑platform, enabling “define once, use many times”.

Self‑service & Intelligent (future): Ongoing efforts to enhance self‑service and AI‑driven analytics.

02 KwaiBI Platform 1.0 Practice

This section details the BI 1.0 implementation that solved early challenges such as high analysis thresholds, low delivery efficiency, and quality assurance difficulties.

2.1 Challenges of the Tool‑Based Era

High analysis threshold: Users needed to master multiple tools, leading to fragmented workflows.

Poor delivery efficiency: Custom reports required at least two weeks per delivery.

Quality assurance difficulty: Lack of standardized development pipelines made stability hard to guarantee.

2.2 Construction Idea for Platform‑Based 1.0

The solution abstracts the platform into a three‑layer architecture:

Data Ingestion Layer: Unified datasets allow flexible connection of heterogeneous sources.

Data Service Layer: A unified analysis query engine abstracts away engine‑specific SQL (HiveSQL, ClickHouseSQL, DruidJSON).

Data Application Layer: Builds various user‑facing analysis tools on top of datasets and services.

This architecture delivers a one‑stop experience and resolves the three major pain points.

2.3 Data Ingestion Layer

Unified datasets evolved through three phases: initial data‑set ecosystem, semantic model for multi‑table relationships, and finally a visual modeling tool enabling drag‑and‑drop data modeling.

Current capabilities: support for dozens of source types, >100k datasets, >20 applications, and tens of millions of data tasks.

2.4 Data Service Layer

Challenges: complex analysis calculations across heterogeneous engines. Solution: a unified DSL (OAX) and a unified query engine that parses OAX, generates execution plans, and performs second‑stage calculations, supporting dozens of functions and million‑level daily queries with sub‑second response.

2.5 Data Analysis Layer

Traditional BI fixed reports hindered secondary analysis. KwaiBI introduced a self‑service analysis tool that enables drag‑and‑drop queries, visualizations, and one‑click sharing to dashboards, mobile, screens, and pushes, with an open SDK for embedding.

The module handles billion‑scale data with second‑level response, and falls back to Hive for ultra‑large datasets.

03 KwaiBI Platform 2.0 Practice – BI Based on Indicator Platform

Building on 1.0, KwaiBI integrates tightly with the indicator middle‑platform to improve data quality and efficiency.

3.1 Quality & Efficiency Challenges

Multiple developers defining the same metric independently caused naming, definition, and output inconsistencies, leading to data quality issues and high maintenance costs.

3.2 Standardized 2.0 – Indicator‑Based BI Solution

KwaiBI adopts a Headless‑BI model: a single definition in the indicator platform, developed by professional data engineers, and consumed across all BI tools, ensuring consistency.

3.3 Architecture Design

Two dataset modes exist: non‑standard UGC (user‑generated) where users define and use their own datasets, and standard PGC (professionally‑generated) where metrics are centrally defined and shared, guaranteeing consistency.

3.4 Summary of 2.0 Practice

Standard datasets dramatically improve data quality, reduce duplication, and increase average query users per dataset from 10 to over 40, with a 30‑fold increase in reuse.

3.5 Business Case in E‑commerce

Professional data engineers develop metrics in the indicator platform, which are then instantly available to all BI tools, boosting analysis efficiency by over 10× and ensuring long‑term quality.

04 Summary and Outlook

KwaiBI combines a flexible UGC data‑set model with a standardized PGC model built on the indicator platform, delivering a unified data service that supports data acquisition, analysis, visualization, and portal creation for the entire company.

4.1 Overall Architecture Review

4.2 Future Outlook

Future work will focus on deeper self‑service capabilities (more comprehensive tools, higher efficiency, better usability) and intelligent features (automated attribution analysis, predictive analytics, smart chart recommendations, and metadata‑driven performance optimization).

05 Q&A

Q: How does Kuaishou’s BI differ from other vendors?

A: Most companies manage metric definitions in an indicator platform but implement downstream BI queries separately, leading to inconsistent definitions. KwaiBI integrates the indicator platform with BI, providing a unified data‑set service that ensures “define once, use many times”.

Q: How is performance handled for detail‑level queries?

A: Three core strategies: ClickHouse projections for materialization, query‑engine optimizations and distributed computation, and fallback to Hive asynchronous queries for massive datasets.

Q: Are Hive data changes synchronized to hot engines in real time?

A: Changes propagate via lineage; field changes trigger automatic notifications for downstream tables.

Q: How are metrics integrated into BI?

A: Metrics, dimensions, and models are registered in a unified data‑set abstraction, exposing a consistent query service to all BI tools.

Q: Is cold‑data sync to hot engines manual or automatic?

A: The data‑set provides an automated acceleration service that routinely imports data to hot engines, improving query performance.