**Estimated reading time:** 12–15 minutes

## Definition

**Data Quality** describes whether data is fit for the business purpose for which it is being used.

High-quality data is not simply data that exists or data that successfully loaded into a database. It must be sufficiently accurate, complete, consistent, timely, valid, and unique for the decisions being made from it.

[[Data Governance]] establishes the expectations and accountability for quality. Data Quality measures whether the data actually meets those expectations.

## Simple Mental Model

> **Data Governance defines what "good" means. Data Quality tells us whether today's data is actually good.**

A pipeline can run successfully and still produce poor-quality data.

```text
Pipeline status: SUCCESS
        ↓
Rows loaded: 1,000,000
        ↓
Half the platform values are NULL
        ↓
Technically successful
Business quality: BAD
```

## How It Works

Common dimensions of Data Quality include:

### Accuracy
Does the data correctly represent reality?

### Completeness
Is required data present?

### Consistency
Is the same concept represented the same way across systems?

### Timeliness
Is the data available when the business needs it?

### Validity
Does the data follow expected rules?

### Uniqueness
Are duplicate records creating false counts?

## Example

Suppose Power BI shows:

> Yesterday's watch hours dropped by 35%.

Before concluding that audience behavior changed, the team checks quality:

```text
Source events received yesterday: 6.4M
Normal daily range: 9M–10M
```

The dashboard calculation may be correct. The real problem is incomplete source data.

## How It Fits Into the Bigger Picture

```text
Source
  ↓
[[ETL vs ELT]]
  ↓
[[Medallion Architecture]]
  ↓
Bronze
  ↓
Silver ← quality checks often become critical here
  ↓
Gold
  ↓
[[Semantic Layer]]
  ↓
Power BI
```

[[Data Lineage]] helps locate where a quality problem entered the chain.

## My Company / Real-World Context

For a media company, useful quality checks could include:

- Daily viewing-event volume
- Percentage of missing content IDs
- Duplicate event rate
- Percentage of unknown platforms
- Missing demographic percentage
- Data refresh completion time
- Difference between source-system totals and governed analytics totals

A high "Unknown" percentage is not automatically poor quality. The key question is whether it is expected due to consent, identity availability, or platform limitations, or caused by a collection problem.

## CTO Perspective

Important datasets can have explicit expectations such as:

```text
Daily refresh completed by 07:00
NULL content_id < 0.5%
Duplicate events < 0.1%
Source-vs-ingested row variance < 2%
```

The objective is not perfect data. The objective is quality appropriate to the business decision and risk.

### Questions to Ask

- What quality dimensions matter for this dataset?
- What thresholds define acceptable quality?
- Who owns those thresholds?
- Are quality checks automated?
- Who gets alerted when a check fails?
- Can we trace the issue through [[Data Lineage]]?
- What business decisions are affected when quality fails?

## Meeting Scenario

**Situation:** An executive says, "The dashboard must be wrong. Audience dropped 30% yesterday."

**Possible response:**

> "Before we interpret this as an audience change, let's validate the quality of the underlying data. I want to confirm source volume, pipeline completeness, and whether any upstream fields changed before we conclude that the business actually moved."

## Key Takeaways

- Data Quality means data is fit for its intended business purpose.
- Accuracy, completeness, consistency, timeliness, validity, and uniqueness are common quality dimensions.
- A successful pipeline can still produce poor-quality data.
- [[Data Governance]] defines expectations; Data Quality measures whether they are met.
- [[Data Lineage]] helps identify where problems originated.

## Related Concepts

- [[Data Governance]]
- [[Data Lineage]]
- [[Data Observability]]
- [[Metadata]]
- [[Data Contract]]
- [[Medallion Architecture]]
- [[Semantic Layer]]
- [[KPI]]