How to Improve Data Quality: A Checklist for Accuracy, Completeness and Freshness
A practical checklist for improving data quality, with accuracy, completeness, freshness and duplicate checks you can run in a spreadsheet or SQL, plus simple monitoring ideas.
To improve data quality, pick the datasets that drive decisions, define what "good" means for each using 6 dimensions (accuracy, completeness, consistency, timeliness, uniqueness and validity), fix problems at the point of entry, and set up simple automated checks with alerts. Assign an owner to every critical dataset and track a few quality scores over time.
Photo: Unsplash
On this page
Bad data rarely announces itself. It shows up as a sales report that doesn’t match finance, a customer who gets the same email 3 times, a dashboard that quietly stopped updating last Tuesday, or an AI tool that gives confident answers built on stale records.
Most businesses respond by cleaning data when someone complains. That works once. Improving data quality means building simple checks and habits so problems are caught early, ideally before anyone sees a wrong number.
This checklist covers how to define good data, the specific checks to run for accuracy, completeness and freshness, and lightweight monitoring that works whether you live in spreadsheets or a cloud data warehouse.
What data quality actually means
Data quality is whether your data is fit for the job you need it to do. The same customer list might be good enough for a newsletter and not good enough for invoicing.
The most widely used framework comes from DAMA UK, the UK chapter of the Data Management Association, which set out 6 dimensions. The UK Government Data Quality Hub recommends the same 6.
| Dimension | Plain-English question | Business example of failure |
|---|---|---|
| Accuracy | Does the data reflect reality? | Customer’s address is outdated, so orders go to the wrong place |
| Completeness | Is all the data we need present? | 30% of leads have no phone number, so sales can’t call them |
| Consistency | Do values agree within and across systems? | CRM says the deal closed; billing has no invoice |
| Timeliness | Is the data available when expected? | Inventory report runs on yesterday’s numbers during a sale |
| Uniqueness | Does each real-world thing appear only once? | Same customer exists 3 times with different spellings |
| Validity | Does data match the expected format, type and range? | Dates stored as text, negative quantities, made-up zip codes |
One subtle point: valid data can still be inaccurate. A perfectly formatted phone number can belong to someone else. Consistent data can be consistently wrong. That is why you need checks on more than one dimension.
Why data quality matters more in 2026
Data quality has always affected reporting. Two trends raise the stakes now.
- AI uses your data directly. Chatbots, copilots and AI agents that read your CRM, product catalog or knowledge base will repeat whatever errors they find. Poor inputs produce confident wrong outputs.
- More automation, less human review. When pricing, reorders, emails and reports run automatically, a bad record can spread to thousands of actions before anyone notices.
Gartner says poor data quality costs organizations at least $12.9 million a year on average, based on its 2020 research, and that many organizations don’t measure data quality at all. Treat that figure as a dated, large-company benchmark rather than your number. Your own cost shows up in returned shipments, wasted marketing spend, staff time spent reconciling reports and decisions made on the wrong numbers.
Key takeaway: You can’t fix every dataset at once. Pick the few that drive money and decisions, define what good looks like for each, and check them automatically.
The data quality checklist
Use this checklist dataset by dataset. Start with 1 critical table, such as customers, orders or products.
1. Scope and ownership
- [ ] List the datasets that feed key decisions, reports or customer-facing systems
- [ ] Name a business owner for each one (the person who decides what “correct” means)
- [ ] Name a technical steward who maintains checks and fixes pipelines
- [ ] Document where each dataset comes from and which systems use it
- [ ] Write down the 5 to 10 fields that matter most in each dataset
2. Accuracy checks
- [ ] Compare a random sample of records against the real source (a contract, invoice or the customer) each quarter
- [ ] Validate mailing addresses against postal reference data; the USPS CASS program certifies address-matching software for US addresses
- [ ] Verify email addresses at signup with confirmation emails
- [ ] Reconcile totals between systems, for example monthly revenue in your store vs your accounting software
- [ ] Flag statistical outliers, such as order values far outside the normal range, for review
3. Completeness checks
- [ ] Measure the percentage of blank values in each critical field
- [ ] Mark which fields are required and which are optional, then enforce required ones at entry
- [ ] Check row counts against expectations (a daily orders table with zero rows is a red flag)
- [ ] Confirm every source system loaded, not just most of them
4. Freshness and timeliness checks
- [ ] Record when each dataset last updated
- [ ] Set an expected update schedule, such as every hour or every morning by 7 a.m.
- [ ] Alert when a dataset is older than its threshold
- [ ] Show “last updated” timestamps on dashboards so users can judge for themselves
5. Uniqueness and consistency checks
- [ ] Test that primary keys, such as order ID or customer ID, are unique and never blank
- [ ] Run duplicate detection on names, emails and phone numbers using normalized values
- [ ] Check that every order points to a customer that exists (no orphan records)
- [ ] Compare key fields across systems, such as customer status in CRM vs billing
6. Validity checks
- [ ] Enforce data types: dates as dates, numbers as numbers
- [ ] Use dropdowns or picklists instead of free text for categories, states and statuses
- [ ] Check values fall within allowed ranges (no negative quantities, no future birth dates)
- [ ] Validate formats for zip codes, phone numbers, SKUs and email addresses
7. Process and prevention
- [ ] Fix the root cause in the source system, not just the report
- [ ] Add validation to web forms and import templates
- [ ] Train staff who enter data on the 3 to 5 rules that matter most
- [ ] Log every fix so you can spot recurring problems
- [ ] Review data quality scores monthly with dataset owners
Simple monitoring ideas that actually work
You don’t need an expensive platform to start. Choose the level that matches your setup.
| Setup | How to monitor | Example tools |
|---|---|---|
| Spreadsheets | Formula-based checks on a “quality” tab, conditional formatting for blanks and duplicates | Excel, Google Sheets |
| Database or warehouse | Scheduled SQL queries that count failures and send an alert if the count is above zero | SQL scripts, scheduled queries |
| Transformation pipelines | Built-in tests on every run | dbt tests and source freshness checks |
| Larger data teams | Data observability platforms that learn normal patterns and flag anomalies | Commercial and open-source observability tools |
Checks you can write today
A useful pattern: each check is a query that returns the bad rows. If it returns nothing, the check passes. This is the same approach dbt uses for its built-in tests, which cover unique values, non-null values, accepted values and relationships between tables. dbt’s documentation recommends that every model has at least a unique and not-null test on its primary key.
Start with these 5 checks on your most important table:
- Primary key is unique. Count IDs that appear more than once.
- Required fields are filled. Count rows where key fields are blank.
- Values are allowed. Count rows where status or category isn’t in the approved list.
- Data is fresh. Compare the newest timestamp with the current time.
- Volume is normal. Compare today’s row count with the average of the last 7 or 30 days.
If you’re new to queries, our SQL for beginners guide covers the COUNT, GROUP BY and WHERE clauses these checks use.
Track a data quality score
Turn checks into a simple score per dataset, such as the percentage of checks passing or the percentage of records that pass all rules. Put the scores on a business dashboard next to the metrics they support. Trends matter more than the exact number: a score that drops suddenly usually means a broken pipeline or a new data entry problem.
Fix problems at the source
Cleaning data downstream is necessary, but it is the most expensive place to fix errors. Each fix gets repeated every time new bad data arrives.
Better places to intervene:
- Forms and apps. Required fields, dropdowns, format validation and address autocomplete stop many errors before they exist.
- Imports. Use a standard template and validate files before they load.
- Integrations. When 2 systems sync, decide which one owns each field so they don’t overwrite each other.
- People. Make it easy to report a data error, and close the loop when it’s fixed.
For one-off cleanup of messy files, follow our step-by-step guide on how to clean data. For recurring problems with duplicate customers or products across many systems, look at master data management.
Where to start: a 30-day plan
Week 1: Pick and profile. Choose 1 critical dataset. Count rows, blanks, duplicates and invalid values. Write down the 5 biggest problems.
Week 2: Define and assign. With the business owner, agree on rules for the key fields and what thresholds are acceptable. Record them in a short shared document or your data governance playbook.
Week 3: Automate checks. Turn the rules into scheduled checks in your spreadsheet, SQL tool or pipeline. Send failures to an email or chat channel someone actually reads.
Week 4: Fix root causes. Tackle the top 2 sources of errors in the systems where data is created. Then repeat the cycle with the next dataset.
Conclusion
Improving data quality is less about a big cleanup and more about steady habits: clear owners, written rules, automated checks and fixes at the source. Use the 6 dimensions to define what good looks like, run the checklist on your most important dataset first, and monitor freshness and duplicates automatically. Within a month you’ll catch problems before they reach a report, a customer or an AI tool.
Sources
- UK Government Data Quality Hub: Meet the data quality dimensions — definitions of the 6 DAMA UK data quality dimensions
- dbt Developer Hub: Add data tests to your DAG — built-in tests for unique, not-null, accepted values and relationships
- dbt Developer Hub: Add sources to your DAG — source freshness thresholds and alerts
- USPS PostalPro: CASS certification — certification of US address-matching software
- Gartner: Data quality — 2020 estimate of the average annual cost of poor data quality
Frequently asked questions
What are the 6 dimensions of data quality?
The 6 dimensions are accuracy, completeness, consistency, timeliness, uniqueness and validity. They come from DAMA UK and are also recommended by the UK Government Data Quality Hub. Accuracy asks if data reflects reality, completeness if needed data is present, consistency if values agree, timeliness if data is available when needed, uniqueness if records appear once and validity if formats and ranges are correct.
How do you measure data quality?
Measure data quality by turning rules into checks and tracking the pass rate. Common measures include the percentage of blank values in key fields, duplicate rate, percentage of records failing format rules, hours since last update and how often totals reconcile between systems. Combine them into a simple score per dataset and watch the trend over time, not just a single number.
What causes poor data quality?
The most common causes are manual entry without validation, free-text fields where dropdowns belong, systems that sync without clear rules about which one owns each field, imports from inconsistent files, broken data pipelines and records that are never updated after customers move or change details. Most problems start where data is created, which is why fixing entry points works better than repeated cleanup.
What is the difference between data cleaning and data quality?
Data cleaning is a one-time or periodic task that fixes errors in a dataset, such as removing duplicates or standardizing formats. Data quality is the ongoing practice of defining standards, monitoring data against them and preventing errors at the source. Cleaning is part of data quality, but without monitoring and prevention, the same problems come back.
Do I need a data quality tool?
Not at first. Small teams can run effective checks with spreadsheet formulas or scheduled SQL queries that count failing rows and send alerts. If you use dbt, its built-in tests and freshness checks cover a lot. Dedicated data quality or observability tools become worth it when you have many pipelines, many data sources or a team that can't manually watch everything.
Who is responsible for data quality?
Data quality is a shared job. A business owner decides what correct looks like for each dataset and sets priorities. A technical steward or data engineer builds checks and fixes pipelines. Everyone who enters data affects quality, so clear entry rules and training matter. Without a named owner, quality problems tend to bounce between teams and never get fixed.
How often should I check data quality?
Check critical data every time it updates, using automated tests that run with each load or sync. Review overall quality scores with dataset owners monthly. Do a deeper manual accuracy audit, comparing a sample of records with real-world sources, each quarter. Freshness checks should run as often as the data is expected to change, which may be hourly for sales and inventory.
Was this guide useful
No reviews yet. Share your experience and help other readers.
Write a review
Manishdai
Manishdai writes about the technology that is reshaping every industry — from AI agents and cloud platforms to healthtech, fintech and the IT companies building it all across the United States.
manishdai.com