15.5 Data Quality & Stewardship

1. Quality is purpose-dependent

There is no universal statement:

"This dataset is high quality."

Instead ask:

"Is this dataset fit for the agreed purpose?"

A planning dataset may tolerate different latency and completeness characteristics than regulatory reporting.

2. Core quality dimensions

Dimension	Example
Completeness	Required business records exist
Validity	Values follow allowed rules
Accuracy	Values represent reality
Consistency	Related systems agree
Timeliness	Data arrives within expectation
Uniqueness	Duplicate records are controlled
Integrity	Relationships remain valid

3. Quality rule design

A useful rule has:

• business meaning,
• technical expression,
• threshold,
• severity,
• owner,
• remediation process.

Example:

Rule:
Every active supplier must have a valid supplier identifier.

Threshold:
100% for Tier 1 supplier master.

Severity:
Critical.

Owner:
Procurement Data Steward.

Action:
Block certification and open incident.

4. Quality dimensions versus pipeline checks

Pipeline checks answer:

"Did the job run?"

Quality checks answer:

"Did the data remain fit for purpose?"

Both are necessary.

5. Quality operating loop

Measure
   |
   v
Detect
   |
   v
Classify
   |
   v
Assign owner
   |
   v
Remediate
   |
   v
Validate
   |
   v
Close

6. Stewardship model

A steward should not become a human ticket router.

Stewards should:

• define and maintain metadata,
• coordinate quality remediation,
• validate definitions,
• monitor governance compliance,
• help consumers,
• escalate systemic issues.

Engineering teams should own technical remediation where appropriate.

7. Quality thresholds

Avoid a single enterprise threshold.

Use criticality:

Tier	Example	Typical posture
Tier 1	Regulatory KPI	Strict
Tier 2	Domain reporting	Controlled
Tier 3	Exploratory	Best effort

Thresholds should be explicit and agreed with consumers.

8. Quality incidents

Classify incidents by business impact:

• informational,
• minor,
• major,
• critical.

A critical quality issue may require:

• consumer notification,
• data product suspension,
• correction,
• backfill,
• executive escalation.

9. Quality and certification

Certification should be a governed decision.

Possible enterprise states:

• Draft,
• Under Review,
• Certified,
• Restricted,
• Deprecated.

These are governance states; do not confuse them with SAP product lifecycle states unless the product documentation explicitly defines the same terminology.

10. Architect exercise

Design a quality framework for an executive Sales data product.

Define:

• five quality rules,
• thresholds,
• ownership,
• alerting,
• consumer notification,
• certification gate,
• remediation SLA.