Skip to main content
Data Quality Dimensions

How to Run a Data Quality Assessment for Any Dataset

A data quality assessment gives you a structured picture of any dataset's strengths and weaknesses. Here's a repeatable process you can run on any data source in 4 hours.

Selva SantoshData quality, for people who ship
2 min read
Key Takeaways
  • Define the intended use before opening the data — requirements determine what counts as acceptable quality
  • Data profiling (null rates, distinct counts, distributions) is the foundation of any assessment
  • Dimension-specific checks: completeness, uniqueness, validity, consistency, timeliness
  • Document findings with severity, affected record count, and estimated remediation effort
  • The assessment output is a priority matrix — not a cleanup project, a diagnostic for one

What a Data Quality Assessment Produces

A data quality assessment answers the question: "Is this data fit for its intended use?" It produces a documented understanding of:

  • Which quality dimensions are strong
  • Which have measurable problems
  • What the likely business impact of the problems is
  • What remediation is recommended

It's not a cleanup project. It's a diagnostic that informs cleanup prioritization.

Step 1: Define the Scope and Intended Use (30 minutes)

Before opening the data, define:

  • What decisions or processes will this data support?
  • What dimensions are most critical for this use case?
  • What are the required fields and their expected completeness thresholds?
  • What accuracy level is needed?

Sohovi profiles every column in your dataset for completeness and flags the exact rows where values are missing — free to try.

A dataset used for bulk email campaigns has different requirements than one used for financial reporting.

Step 2: Data Profiling (1–2 hours)

Run basic profiling on the dataset:

  • Row count (how many records?)
  • Column-level null rates (what % of each column is null?)
  • Distinct value counts (how many unique values per column?)
  • Value distribution for key columns (are values evenly distributed or concentrated?)
  • Min, max, mean for numeric columns (are ranges plausible?)
  • Sample values for text columns (spot-check for obvious errors)

Tools: Python pandas, OpenRefine, Excel pivot tables, or a SQL query against the table.

Step 3: Dimension-Specific Checks (1–2 hours)

Completeness: Which columns fall below required thresholds? Uniqueness: Does the primary key have duplicates? What's the duplicate rate? Validity: Do values in key columns pass format and range checks? Consistency: For datasets with relationships, are foreign keys intact? Do totals match their components? Timeliness: When was the data last updated? Is that recent enough for the intended use?

Sohovi automatically finds every duplicate in your dataset — including near-matches — and shows you exactly which rows are affected.

Step 4: Document Findings (30 minutes)

For each issue found:

  • Dimension affected
  • Field(s) affected
  • Severity (critical / significant / minor)
  • Count and % of records affected
  • Recommended remediation
  • Estimated remediation effort

Step 5: Summarize and Prioritize

Overall fitness assessment: Is this data usable for its intended purpose as-is? With remediation? Not at all until significant cleanup?

Priority ranking of findings by business impact.

A two-page findings summary with a priority matrix is the typical deliverable.

Frequently Asked Questions

What's the fastest way to profile a CSV dataset?

Python pandas: df.describe() for numeric columns, df.isnull().sum() for null counts, df.nunique() for distinct value counts. For 100k rows, this takes seconds and gives you most of what you need for an initial profile.

How do I assess accuracy during a data quality assessment?

Spot verification: sample 20–50 records and manually verify key fields against an authoritative source. This gives you an accuracy estimate. Full accuracy verification at scale requires external enrichment tools.

Who should I involve in a data quality assessment?

The data owner (knows what the data should contain), the primary user (knows what decisions it supports), and ideally someone with the technical skill to run the profiling queries. For most assessments, 2–3 people are sufficient.

Selva Santosh

Data quality, for people who ship

Selva writes practical guides on data quality, profiling, and governance to help teams ship better data.

Start for free

Stop guessing. Start knowing your data quality.

Sohovi profiles your datasets in minutes — surfacing completeness gaps, type mismatches, and duplicate patterns before they reach production.

Try Sohovi freeMore articles

No credit card required · Free forever plan

More from Data Quality Dimensions

All posts →
Data Quality Dimensions

The 6 Dimensions of Data Quality, Explained with One Messy Spreadsheet

9 min read · Jun 12, 2026

Data Drift vs Schema Drift: What's the Difference?Data Quality Dimensions

Data Drift vs Schema Drift: What's the Difference?

7 min read · Jun 11, 2026

How Z-Scores Are Used in Data Quality MonitoringData Quality Dimensions

How Z-Scores Are Used in Data Quality Monitoring

7 min read · Jun 11, 2026

How to Run a Data Quality Assessment for Any Dataset | Sohovi