Consolidating 14 source systems for a top-10 Vietnamese bank

Customer context

A top-10 Vietnamese joint stock commercial bank with more than 200 branches and transaction offices. After years of expansion and acquiring systems from multiple vendors, business data sat across 14 systems: core banking, cards, lending, treasury, CRM, digital channels and several internally built applications.

The challenge

Three problems the executive team named as priorities:

• A 12-working-day financial close. Most of it spent reconciling numbers between systems by hand through intermediate spreadsheets.
• Inconsistent numbers across divisions. “Closing loan balance” was computed three different ways by three divisions, differing by 2–5% and consuming every review meeting.
• Analytical queries loading the transactional core. Ad-hoc reports ran directly against core banking, slowing transactions during month-end peaks.

What BigAI delivered

Four phases, each with its own acceptance criteria so the bank saw value early rather than waiting until the end.

Phase 1 — Assessment and design (6 weeks)

Mapped all 14 source systems and interviewed nine business divisions to agree definitions for 260 core metrics. The most important output was not a technical document but a metric dictionary signed off by every division.

Phase 2 — Lakehouse platform (10 weeks)

Built a Bronze/Silver/Gold lakehouse on Apache Iceberg, running entirely on the bank’s own infrastructure. Core banking data loaded via CDC to protect the source system; the rest via nightly batch.

Phase 3 — Reconciliation automation (6 weeks)

Converted 34 manual reconciliation procedures into automated pipelines with data quality rules. Discrepancies now surface the same day instead of at close.

Phase 4 — Handover and training (4 weeks)

Transferred ownership to a seven-person internal data team with operating documentation, incident runbooks and five hands-on training sessions, plus six months of support.

Results

Swipe to see the full table

The biggest value was not the technology. It was that, for the first time, the divisions sat down and agreed on a shared definition for every metric. Everything after that was execution.

Technology used