AI-Driven BI: Achieving Zero-Barrier Data Access and Smart Insights

Background and Trends

In recent years, BI platforms have increasingly merged with AI, focusing on pure business users and leveraging natural language processing to build search‑driven data analysis platforms that lower the barrier to data consumption.

BI Platform Evolution

The BI concept was first introduced by Gartner in 1996 as a fact‑based decision‑support system, initially centered on IT‑driven predefined reporting. Over two decades, traditional BI grew stronger but remained limited to fixed summary reports, leading to long response cycles and low value visibility.

In the early 2000s, self‑service data analysis platforms for business users emerged, giving rise to “agile BI”. While agile BI improved collaboration, it struggled with long‑tail data processing, knowledge retention, and sustained IT support.

Since 2010, especially in the past five years, BI has integrated with AI, emphasizing business‑centric design, natural language interfaces, and zero‑threshold data consumption, while also supporting long‑tail data, knowledge accumulation, and experience sharing.

TikTok Group Enterprise BI Journey

From 2018 to present, TikTok’s internal data analysis platform has undergone three stages: (1) solidifying foundations to improve data extraction and analysis efficiency; (2) enriching capabilities with multi‑terminal, enterprise management, and intelligent attribution features; (3) innovating with visual modeling for drag‑and‑drop data cleaning and exploring large‑model integration for analysis assistants.

The platform’s intelligent data insight architecture consists of five layers—data ingestion, query engine, data modeling, data analysis, and data presentation—providing high operability and supporting massive data processing with sub‑second query latency.

High‑Performance Data Analysis Architecture

A robust big‑data architecture and high‑performance query engine are the foundation of efficient analysis. The system uses ByteHouse, a cloud‑native, deeply optimized version of ClickHouse, to accelerate queries and implements a deep extraction‑query chain.

Data is stored exclusively in ByteHouse without redundancy, enabling shared access across services and direct connections for custom queries, ensuring data accessibility and sovereignty.

Application‑Level Optimization and Governance

Beyond engine performance, fine‑grained application‑level tuning is essential. Techniques include multi‑level aggregation, hot‑cold data tiering, and intelligent routing that selects the most efficient aggregation table or falls back to wide tables or Presto queries.

Performance governance ensures query p95 stays under two seconds through cluster isolation, resource groups, automatic task parameter optimization, and proactive recommendations for primary keys, sharding, and indexing.

BI + AI for Intelligent Data Insight

AI capabilities enhance data mining, attribution analysis, and decision support. The platform offers visual modeling with over 30 built‑in operators, enabling users to drag, configure, and apply common AI operators such as classification, regression, and prediction.

Pre‑built AI models, like ARIMA for ad spend forecasting, can be trained on user data and integrated into workflows for predictive insights.

AI‑Powered Data Mining and Modeling

AI‑driven attribution includes real‑time anomaly detection, dimension attribution, and metric attribution, generating automated reports that can be pushed via instant messaging for rapid response.

The system supports custom attribution configurations, scheduled analyses, and multi‑dimensional drill‑downs, with visualizations illustrating contribution of cities, provinces, and channels to metric changes.

Large‑Model Assisted Analysis

Following the rise of ChatGPT, large models are applied to lower the skill barrier for data analysis. DataWind’s analysis assistant leverages LLMs for natural‑language SQL generation, dataset description, and chart creation, as well as embedded AI assistants for collaborative workflows.

Use cases span data preparation (auto‑generating metadata and SQL assistance), self‑service analysis (NL‑driven chart generation), and dashboard consumption (custom themes and IM‑based exploration).

Conclusion and Outlook

Gartner’s latest Magic Quadrant adds metrics store, collaboration, and data‑science integration as key BI capabilities, aligning with trends toward high‑performance data consumption and AI‑driven intelligence.

Future directions include accelerating data processing, lowering analysis barriers, and embedding AI assistants across data pipelines, from ingestion to collaborative consumption.