เงินเทอร์โบ · Ngernturbo · Data & Platform

Trust in Every Number

In May we put a data quality framework in front of the warehouse. Three months later the warehouse is 10× cleaner — and bad data is stopped before it lands.

Executive briefing · August 2026 Data Engineering team
Click anywhere or press Enter ⏎ to begin
The change · May 2026

From invisible problems to a daily checkpoint.

Before

sources
warehouse
reports
wrong number spotted here
  • No checkpoint — pipelines trusted whatever arrived from source systems.
  • Found by the business — a user reading a wrong report was the first detector, weeks after the fact.
  • No evidence — no record of what was checked or what failed; every investigation started from zero.

After — live since 21 May 2026

  • Our own product: ntb-dq-framework, a Python package built in-house and published on PyPI — with its own AI assistant for rule authoring.
  • Six quality dimensions, checked on every load — and a named owner behind every rule.
100%
Days on schedule
250+
Active rules
39
Tables covered
70M+
Rows checked
The six dimensions

Six questions we ask of every table.

Completeness

Is anything missing?

Required fields that arrive empty, expected date partitions that never landed, and row counts that suddenly drop against their baseline.

Consistency

Does it agree with itself?

References that point to nothing, values that disagree between systems, and business logic that must always hold — an end date can never come before its start date.

Uniqueness

Is anything counted twice?

Duplicate keys that would silently double-count loans, customers, or payments in every report built on top.

Validity

Is every value well-formed?

ID cards that must be exactly 13 digits, values outside an allowed list or a sane range, and data of the wrong type.

Timeliness

Is the data fresh?

Hours of delay since a feed last updated, and alerts when a late feed breaches its agreed SLA.

Accuracy

Does it match the source of truth?

Row-level and total reconciliation against the core bank — the next dimension we switch on.

How it works today

Every load takes one of three paths.

incoming load
250+ active rules · checked daily
Pass
target table
99.997% of all row-checks — clean data flows through untouched
Warning
target table + DQ log
12K+ rows logged — the row still lands, by agreement with data owners; the trend is watched
Critical
Quarantine zone
3,700+ rows stopped — never reaching the warehouse

Quarantine is reviewed jointly: the data engineer diagnoses the pipeline side; the data owner rules on the business side.

The impact · three months in
10×
May — baseline June — down 74% July — down 87% August — down 90% May Aug

cleaner warehouse in three months

100%
days on schedule
99.998%
of checks pass today
100%
critical failures quarantined

Duplicate loan contracts, broken loads, malformed ID cards — every one caught the same day, with the evidence saved.

Trust in every number.

Built in-house by the Data Engineering team · เงินเทอร์โบ

เงินเทอร์โบ · Data & Platform
1 / 5