When you ask "how is our data?" inside an enterprise, you get three unrelated numbers from three teams. Data engineering talks about indirect table-level tests; the business asks "are reports correct"; the CDO wants a single KPI. Trust Score collapses those three views into one metric.
What is the Trust Score?
The Trust Score expresses, on a 0-100 scale and over a defined time window, how trustworthy a data asset (table, pipeline, KPI) is. It is composed — not a single test — and built from a weighted combination of six measurable dimensions.
Components
| Dimension | Measurement | Typical weight | |---|---|---| | Accuracy | Records matching a verifiable source | 25% | | Completeness | Required-field population rate | 15% | | Consistency | Cross-source match rate | 15% | | Timeliness | Within-SLA arrival rate (p95 latency) | 15% | | Uniqueness | 1 − duplication rate | 10% | | Validity | Absence of schema/regex/range violations | 20% |
Weights shift by asset type: regulatory layers favour accuracy + timeliness; analytical layers favour consistency.
Formula (example)
TrustScore = 0.25 × Accuracy + 0.20 × Validity + 0.15 × Completeness + 0.15 × Consistency + 0.15 × Timeliness + 0.10 × Uniqueness
Each component is normalised to 0-100. Colour thresholds: 90+ green, 75-89 amber, below 75 red.
Asset-level, not table-level
Trust Score is reported at the asset level, not just the table level. A KPI may feed from multiple tables; the KPI’s Trust Score is the weighted average of its source tables. Combined with lineage, this rolls up automatically and the business talks in terms of one number per KPI.
Threshold breach = incident
When Trust Score drops below the threshold, an incident opens automatically:
- Red (<75): immediate PagerDuty/Slack alert to owner and steward, escalation timer.
- Amber (75-89): visible on the monthly dashboard, owner pulled into a weekly review.
- Green (90+): within SLA, no action required.
Coupled with incident management, "the data is bad" becomes "Customer table at Trust 72 — red, owner Mehmet, SLA 24h".
What it means for the CFO and audit
Trust Score moves the conversation from abstract "data quality" to a trust indicator for operational and financial decisions:
- The CFO writes "production data locked at Trust 91" in the annual report.
- A BDDK audit receives the 30-day history of the regulatory warehouse Trust Score.
- A new AI project sets a Trust Score threshold (e.g. ≥85) on its source assets before kicking off.
Operational discipline
For the programme to be sustainable in production, three controls are mandatory:
- Automation: measurements recompute on every run via dbt + Soda Core + custom tests.
- Ownership: every asset has an owner and a steward; threshold breaches ring both phones.
- Monthly executive review: the CDO dashboard shows the five worst tables, trend, incident summary.
Closing
"Data quality" is the area where everyone agrees on the language and disagrees on the numbers. Trust Score turns the language into a number, the number into an incident, and the incident into something owned and timed. In a mature programme, debates shift from "good or bad" to "Customer table at 92% accuracy, 88% completeness, Trust 87".
