Modernizing Data Governance at First Thai Bank to Prepare for an AI-First Future

The case offers a pragmatic blueprint for enterprises, especially in financial services, seeking to operationalize AI at scale. Organizations navigating similar complexity will find that data governance is the essential foundation before any AI initiative can deliver reliable, auditable results.

SCB moved from fragmented silos to a single governed platform, modernized data quality by bringing validation logic to where data lives, and unlocked self-service discovery through conversational metadata. The result is faster operations, lower cost, and a foundation that can support both regulatory rigor and new AI initiatives.

Unifying a fragmented group on one governed platform

SCB's 2022 shift to SCBX split core banking from credit cards, auto loans, investment platforms, and venture units. Twelve companies meant twelve stacks, inconsistent definitions, and complex data sharing. The bank set goals to eliminate silos, avoid vendor lock-in, standardize critical concepts like "customer," and share data securely across subsidiaries.

The team consolidated legacy warehouse and lake assets into a Unified Data Platform on the Databricks Platform using a medallion architecture (raw, persist, curate). Unity Catalog provides centralized governance and access control for 20 plus business units. Delta Sharing enables secure, cross-tenant and cross-cloud data exchange across the group. More than 10,000 legacy jobs were migrated, 2023 focused on ML use cases, and 2024 expanded to unify EDW and lake while pulling from both old and new core banking systems.

Rebuilding data quality where the data lives

SCB's prior data quality approach copied data out to a separate tool, creating silos and manual rule management for each table. The bank replaced it with a parameter-driven PySpark pipeline on the Databricks Platform. Data stays in Unity Catalog, quality scores are written back to governed tables, and scorecards are shared via Delta Sharing to Power BI. Stewards can drill from overall scores down to source systems and critical data elements.

Performance and cost gains were material. A monthly retail customer job fell from 60 hours to 6 through native execution, right-sized compute, and a staging pattern that avoided end-to-end failures. Daily jobs dropped from 8 hours to about 30 minutes. Eliminating third-party licenses and vendor-specific expertise, combined with faster runtimes, delivered roughly 15 times lower cost. The framework now enforces around 120 rules across 18 million customers while maintaining long-standing scorecard views.

Conversational metadata to accelerate safe discovery

After consolidating and governing data, the bank targeted democratization. Discovery had relied on portal searches, monthly steward emails, and tribal knowledge, and users could not see assets they lacked access to. A Genie-based application now lets users query curated metadata in natural language. It reads three Unity Catalog tables, covering schema, table descriptions, and column descriptions, and runs on Databricks SQL for responsive answers.

Metadata remains curated by an administrator who maintains an Excel source synced via Azure Data Factory and a Lakeflow Jobs job into the three Unity Catalog tables. The payoff is significant. A regulator's request to locate where national ID numbers reside, once a multi-week effort, can return a live inventory of 1,009 attributes in minutes. A data scientist building a single-customer view surfaced 48 customer ID fields across 26 systems in one session, complete with sensitivity labels and system context.

AI acceleration starts with shared, governed context

SCB's experience shows that AI impact follows disciplined platform work: unify silos into one governed lakehouse, run data quality inside the platform, and turn institutional metadata into searchable knowledge. Operational gains in time and cost pair with strategic agility to answer regulators, share across subsidiaries, and launch new AI projects without rebuilding the plumbing.