AI‑Driven Data Governance as a Service: Tencent Games' Paradigm Shift

With the rapid global expansion of Tencent Games and explosive data growth, traditional data governance faced resource waste, low collaboration efficiency, and passive decision‑making. The core goal is to improve data pliability and overall resource efficiency to support stable, iterative business operations.

Data Service Evolution

Initial custom game support stage.

Control stage resembling data‑mid‑platform construction, focusing on data‑table and task management.

Steady‑state stage emphasizing method and tool consolidation, achieving 10% core data asset coverage of 80% custom development needs.

Current AI‑optimization stage, redesigning the delivery pipeline with AI to shorten chains and accelerate updates.

Challenges in the Steady‑State Stage

Passive resource and cost control: governance reacts to issues rather than proactively planning.

Severe storage waste: 30% of historical partitions remain unused, and 20% of compute resources are consumed by obsolete tasks.

Collaboration inefficiency: metadata, warehouse standards, and metric definitions rely heavily on manual maintenance, causing misalignments and high “mouth‑to‑mouth” alignment costs.

AI‑Powered Governance Framework

Metadata semantic learning: AI models ingest historical SQL logs and table schemas to generate human‑readable, machine‑callable semantic tags.

Intelligent demand‑chain reversal: AI agents use technical lineage (task‑table relationships) to infer missing or obsolete business requirements, producing "SQL‑to‑Text" mappings.

Unified semantic model: a three‑layer knowledge hierarchy (general knowledge from large models, industry‑specific gaming terminology, and internal corporate jargon) ensures consistent cross‑team understanding.

Service Scenarios

Resource intelligent control : Transform raw billing data into business‑friendly reports, e.g., linking new feature ABC to 10 new tasks, 5 tables, and a 15% storage cost increase, replacing passive cost queries.

Collaboration efficiency : Deploy a semantic‑driven Q&A system that answers metric‑definition queries with underlying SQL, cost, and reuse information, cutting alignment time by 50%.

Demand development assistance : Match new requests (e.g., holiday event statistics) with high‑quality reusable assets, auto‑generating SQL and reducing development time from 1 hour to 30 minutes.

Results

Manual maintenance workload reduced by ~30% after semantic coverage.

Asset recommendation Top‑3 accuracy reached 95%.

Business development speed improved by 50%; average demand implementation time fell from 1 hour to 0.5 hour.

Completed semantic processing for 5,500 tables and 100,000 fields, establishing an automated asset‑quality evaluation system.

The AI‑driven “governance‑as‑service” model demonstrates that data governance can shift from a cost center to an active service, delivering tangible efficiency gains and enabling data‑centric innovation in the gaming industry and other data‑intensive domains.