A data quality audit is the process of systematically evaluating a dataset to identify specific quality problems — completeness gaps, duplicate records, format errors, and consistency issues — so you can fix them before they damage your operations, decisions, or customer relationships.
If you've never run a data quality audit before, the term can sound technical and intimidating. It isn't. It's a structured review process that most teams can complete in a few hours, and it usually surfaces problems that have been silently costing money for months.
This guide walks you through every step — from choosing the right dataset to auditing to taking action on what you find.
In this guide
- Step 1: Choose which dataset to audit first
- Step 2: Define what "good quality" looks like for this dataset
- Step 3: Export the data and prepare it for review
- Step 4: Run the audit across five quality dimensions
- Step 5: Score and prioritize your findings
- Step 6: Document and act on the results
- What to do with audit findings
Sohovi measures all 10 data quality dimensions — completeness, validity, uniqueness, accuracy, consistency, and more — automatically across every column.
Step 1: Choose Which Dataset to Audit First
Don't try to audit everything at once. Start with the dataset that creates the most risk if it's wrong, or causes the most daily friction when it's messy.
Good candidates for a first audit:
- Your customer contact list — if bad data is reaching customers or your email campaigns are underperforming
- Your CRM data — if sales teams are working from unreliable pipeline data
- Your product catalog — if product listings have inconsistencies that affect sales or inventory
- Your financial records export — if reconciliation takes hours each month
The right dataset to audit first is the one where fixing quality problems will have the most immediate, visible impact on your business.
Step 2: Define What "Good Quality" Looks Like
Before you run the audit, define the standards you're auditing against. Without this step, you'll find problems but won't know which ones matter.
For each key field in the dataset, write down:
- Required completeness: What percentage of rows must have a value? (Email: 98%, phone: 90%, company: 85%)
- Valid format: What does a valid value look like? (Email: contains @, phone: 10 digits, date: YYYY-MM-DD)
- Uniqueness requirement: Should this field be unique? (Customer ID: yes, email: yes with exceptions)
- Allowed values: For categorical fields, what values are acceptable? (Status: Active / Inactive / Pending)
Sohovi profiles every column in your dataset for completeness and flags the exact rows where values are missing — free to try.
Write these down before you start. They become your benchmark — the standard every finding will be measured against.
Step 3: Export the Data and Prepare It for Review
For most businesses, the audit starts with a CSV export from your CRM, database, email platform, or spreadsheet system.
Before you start reviewing, check the basics:
- Row count: does the number match what you'd expect from the source system?
- Column count: are all expected fields present?
- Header row: are column names correct and consistent?
- Encoding: open in a text editor — are there garbled characters in any fields?
These structural checks take 5 minutes and catch problems that would corrupt everything downstream.
Uploading to Sohovi gives you an instant structural and statistical overview — completeness rates per column, duplicate counts, and format patterns — before you begin the manual review. It's a good first pass that tells you where to focus.
Sohovi automatically finds every duplicate in your dataset — including near-matches — and shows you exactly which rows are affected.
Step 4: Run the Audit Across Five Quality Dimensions
Dimension 1: Completeness
What to check: For each required field, what percentage of rows have a value?
How to check: Count null/empty values per column. In a spreadsheet: =COUNTBLANK(column_range)/COUNT(all_rows). In a data quality tool, this is automated.
What to flag: Any required field below your defined threshold. Document: which field, what the null rate is, and whether nulls are truly missing or are placeholder values ("N/A", "0", "unknown").
Dimension 2: Uniqueness
What to check: Are there duplicate records for entities that should be unique?
How to check: For your unique identifier fields (customer ID, email, order number): count how many values appear more than once. In a spreadsheet: =COUNTIF(column, cell)>1 applied across all rows.
What to flag: Any duplicated unique identifier, and any cases where the same name/email appears with different contact details across two records.
Dimension 3: Validity
What to check: Do field values match the expected format?
How to check: For emails: look for addresses without "@", multiple "@" symbols, or no domain. For phone numbers: look for non-numeric characters, too-short or too-long numbers. For dates: look for values that don't parse as a date, or dates outside a plausible range (e.g., birth year = 1800).
What to flag: Any field with a validity failure rate above 1–2%. Document specific examples of invalid values and estimated prevalence.
Dimension 4: Consistency
What to check: Is the same information stored the same way across the dataset?
How to check: Look for mixed formats in the same field: "New York" and "NY" in a state field. "US" and "USA" and "United States" in a country field. "Active" and "active" and "ACTIVE" in a status field.
What to flag: Any field with more than one way of expressing the same value. This is especially important for fields you'll use for filtering, segmentation, or joining to another dataset.
Dimension 5: Timeliness
What to check: Is the data current?
How to check: Look at date fields — last updated, last contacted, created at. What percentage of records haven't been touched in over 12 months? 24 months?
What to flag: Any stale record rate that would affect how you use the data. For a marketing list, a 30% rate of contacts not updated in 18 months is a significant deliverability risk.
Step 5: Score and Prioritize Your Findings
Once you've checked all five dimensions, you'll have a list of findings. Before you try to fix everything, prioritize.
High priority:
- Any finding that affects data currently being used for a business-critical purpose (current campaign list, active pipeline, financial reporting)
- Any finding that creates compliance risk (PII in unexpected fields, sensitive data without proper access controls)
- Any finding where the problem is systemic (not a few errors, but a structural data quality failure)
Medium priority:
- Moderate completeness gaps on important but non-critical fields
- Consistency issues that won't affect current use but will cause problems if the data is joined to another system
Low priority:
- Minor formatting inconsistencies in fields not currently used
- Stale records in segments you won't target in the near term
A simple scoring approach: rate each finding on a scale of 1–3 for both impact (how much damage if not fixed) and ease of fix (how quickly can this be resolved). Fix high-impact, easy-fix items first.
Step 6: Document and Act on the Results
What to document:
- Date of the audit and which dataset was audited
- Each finding: field, dimension, failure rate, specific examples
- Priority rating for each finding
- Recommended action and owner for each item
Acting on findings:
For completeness gaps — investigate why fields are empty. Is it a form that doesn't require the field? A data entry process that skips it? An import that doesn't map it? The fix depends on the cause.
For duplicates — merge or delete duplicate records. For complex cases, investigate which record has the most complete and recent data before deciding which to keep.
For validity failures — clean the bad values by standardizing formats. Set up validation rules going forward to prevent new invalid values from entering.
For consistency issues — standardize the field across all records, then enforce a controlled vocabulary going forward.
For timeliness issues — mark stale records, reach out to verify current information, or flag as lower-confidence data.
Frequently Asked Questions
Q: What is a data quality audit? A data quality audit is a systematic evaluation of a dataset to identify specific quality problems — completeness gaps, duplicate records, format errors, and consistency issues. It produces a prioritized list of findings that tell you what's wrong, how bad it is, and what to do about it.
Q: How long does a data quality audit take? For a typical small business dataset (5,000–50,000 rows), a first audit takes 2–4 hours if done manually. Using an automated tool for the statistical analysis (completeness rates, duplicate counts, format patterns) can cut that to 30–60 minutes for the automated portion, plus time for manual review of specific findings.
Q: How often should you run a data quality audit? For your most important datasets, quarterly audits are a good baseline. For data that changes frequently (email marketing lists, active CRM data), consider a lighter-weight audit before every major use. For stable reference data, annual audits may be sufficient.
Q: What's the difference between a data quality audit and a data quality checklist? A checklist is a routine pre-use review — fast, focused, and run frequently. An audit is a deeper periodic assessment that looks at the overall health of a dataset across all quality dimensions. The checklist catches problems in the moment; the audit catches patterns and systemic issues.
Q: Do I need technical skills to run a data quality audit? No. The statistical analysis (completeness rates, duplicate counts) can be done with spreadsheet formulas or automated tools. The qualitative review (is this field being used correctly? are these values making sense?) requires business knowledge, not technical skills.
Q: What should I do if I find PII in an unexpected place during an audit? Stop, document the finding, and notify your data owner or compliance lead immediately. Unexpected PII in a dataset (SSNs in a notes field, credit card numbers in a description field) is a compliance risk that needs to be addressed before the audit continues on other dimensions.
Q: How do I know when a finding is serious enough to act on immediately? Prioritize based on impact: is this data being used right now for a business-critical purpose? Does the problem create compliance risk? Is it systemic (affecting a large percentage of records) or isolated? High-impact systemic findings should pause the downstream use of the data until they're resolved.
Q: Can I run a data quality audit on just part of a dataset? Yes. If your dataset is very large, auditing a random sample (5–10% of rows) gives you a statistically meaningful picture of quality issues across the full dataset. For fields with known quality problems, you may want to review all records in that field rather than a sample.
Q: What's the most common thing people find during their first data quality audit? Duplicate records and missing values in fields they thought were well-maintained. Most teams are surprised by how many more duplicates exist than they expected, and by which fields have the highest null rates. The first audit almost always reveals problems that have been invisible for months.
Q: What should I do after the audit to prevent the same problems from recurring? After fixing the immediate findings, address the root causes. Add validation rules at data entry points that produced invalid values. Require fields that were optional but critical. Create a recurring audit schedule to catch new problems before they accumulate. Update your data quality policy to reflect what you learned.
Your first data quality audit will almost certainly surface problems you didn't know existed. That's the point. The audit isn't a judgment — it's a map. Once you can see the problems clearly, fixing them is straightforward.
If you're ready to run your first data quality audit, Sohovi makes the statistical analysis instant. Upload your CSV and get a complete quality breakdown — completeness rates, duplicate counts, format issues, potential PII — in under a minute. No credit card, no IT team, no code required.