Intelligent Business Intelligence at Kuaishou: Architecture, Challenges, and Solutions

Background : Kuaishou Data Platform aims to improve decision‑making efficiency by building advanced compute engines and high‑performance data services, offering a comprehensive BI toolchain that supports everything from data ingestion to advanced analytics.

BI Evolution : The platform has progressed through three stages—traditional BI (reporting), agile BI (self‑service and visualization), and intelligent BI (AI‑enabled analytics that lower the barrier for all users).

Current BI Architecture : The system consists of three layers—business, product, and service—covering data sources, standardized metrics, and API services, with AI integration providing intelligent search, SQL generation, and conversational data retrieval.

Challenges : Diverse user requirements, massive and heterogeneous data, high‑cost implementation, and limitations of generic large models (hallucination, low accuracy) hinder intelligent BI adoption.

Solution Approach : Kuaishou proposes a three‑fold strategy—process controllability, result trustworthiness, and feasible deployment—by enhancing product, data, platform, and model layers, including context‑aware multi‑turn dialogue, metadata quality control, fine‑grained workflow orchestration, and vertical domain model training.

Intelligent Service Architecture : A "three‑horizontal, one‑vertical" design comprising infrastructure (large models, vector DB), business framework (agents, workflow orchestration), interface layer (HTTP/RPC APIs), and surrounding ecosystem (algorithm libraries, quality assessment).

Application Practice : Smart BI has been deployed in over ten scenarios, such as SQL auto‑completion, conversational query, attribution analysis, and summary generation, achieving roughly 10% overall SQL generation rate and >20% adoption for individual users.

Future Outlook : Continued integration of BI and AI to deepen analytical capabilities, close the intelligent analysis loop, and further boost user efficiency and decision quality.

Q&A : Includes statistics on SQL generation adoption, strategies to avoid query errors, and ongoing optimization directions.