Data Quality Checklist for Better Decisions
Bad data quietly corrupts dashboards, AI models, and CRM records. This practical checklist covers the six dimensions of data quality, a step-by-step audit, and a fix plan that keeps your data clean.
The Short Answer
Bad data rarely announces itself. It hides inside dashboards that look polished and reports that look confident, until a decision built on them goes wrong. This data quality checklist gives your team a practical way to find those problems early, fix them at the source, and keep your data trustworthy.
Bad data rarely announces itself. It hides inside dashboards that look polished and reports that look confident, until a decision built on them goes wrong. This data quality checklist gives your team a practical way to find those problems early, fix them at the source, and keep your data trustworthy.
Most teams discover a data problem only after the damage is done. A sales forecast misses because the pipeline was built on duplicate accounts. A campaign targets people who left their jobs months ago. An AI tool recommends the wrong product because it learned from messy records.
The scale is well documented. A 2026 OneStream survey of more than 350 senior finance and technology leaders found that 47 percent had made a material business decision in the past year based on inaccurate, incomplete, or outdated financial data. Salesforce research from 2025 estimated that leaders consider about 26 percent of their organisation's data untrustworthy, roughly one record in four. The good news is that data quality is fixable with a clear checklist, a regular audit habit, and someone who owns the result.
Why a Data Quality Checklist Matters in 2026
Businesses now make more decisions from data than ever before, and they make them faster. Dashboards update in real time, AI assistants draft recommendations, and automated workflows act without a human in the loop. That speed is powerful when the data is sound, and dangerous when it is not.
Clean data matters more because AI has raised the stakes. AI tools learn patterns from whatever you feed them, so duplicates and stale contacts become facts the model trusts. A 2025 IBM study of 1,700 chief data officers found that only 26 percent were confident their data could support new AI-driven revenue goals.
For a broader view, see our guide to real-time data analytics in 2026 and how AI-powered analytics help businesses make smarter decisions.
The Six Dimensions of This Data Quality Checklist
Every data quality checklist should test the same six dimensions, drawn from the DAMA Data Management Body of Knowledge. Each catches a different failure, and a record can pass five and still mislead you if the sixth is broken.
| Dimension | What it checks | Typical failure | How to test it |
|---|---|---|---|
| Accuracy | Does the value match reality? | A contact's job title is two years out of date | Sample records and verify them against a trusted source |
| Completeness | Are all required fields filled in? | Four in ten accounts have no industry listed, so segmentation drops them | Measure field fill rates for every required field |
| Consistency | Does the value agree across systems? | Your CRM says a customer churned, your billing system says active | Compare the same field across connected systems |
| Timeliness | Is the data current enough to act on? | A lead list still holds contacts who changed jobs months ago | Check last-verified dates and record age |
| Validity | Does the value follow the required format and rules? | Phone numbers in five formats break dialling and matching | Run format and range rules on every field |
| Uniqueness | Does each real-world entity appear only once? | One customer exists as three accounts with split history | Run duplicate detection on names, emails, and domains |
Accuracy and completeness
Accuracy is the hardest dimension to test automatically, because no rule can tell you whether a job title is true. The practical method is sampling: check a random sample of 200 to 500 records against a trusted source. Completeness is the easiest to measure: calculate the percentage of records where each required field is filled.
Consistency and timeliness
Consistency only exists between systems, never inside one. Compare fields that live in more than one place, such as customer status, across your CRM, billing, and support tools. Mismatches usually point to a broken integration. Timeliness matters because data decays: B2B contact databases decay by roughly 22.5 percent a year, so about one record in four becomes inaccurate within twelve months without refreshes.
Validity and uniqueness
Validity checks whether values conform to defined formats and rules, such as valid email addresses and phone numbers in one standard format. These failures are the easiest to prevent, because validation rules can block bad values at entry. Uniqueness is about duplicates, which split one customer into several partial histories. Industry reports show duplicates account for 10 to 30 percent of contacts in many enterprise CRM environments.
How to Audit Your Data: A Step-by-Step Data Quality Checklist
An audit sounds like a large project, but a first pass can be done in days. Work through the steps below in order and record your findings. The output is a baseline you can compare against after every fix.
Step 1: Map your sources and pick the fields that matter
List every system that holds business data: CRM, billing, marketing automation, support desk, spreadsheets, and legacy imports. Note who owns each one and what it feeds. Then pick the fields that drive decisions, such as customer identity, contact details, deal stage and value, and region, and write down what good looks like for each.
Step 2: Profile your data
Data profiling means scanning a dataset to learn its shape: how many records exist, how many fields are empty, and where the outliers sit. A spreadsheet export and a few summary counts will reveal most problems on a first pass.
Step 3: Detect duplicates and stale records
Start with exact matches on email address, then fuzzy matching on company name and domain, and finally composite matching on name plus phone number or address. Always review merges before applying them, because an aggressive merge can destroy history. Then sort records by last activity. Anything untouched for a year or more is a candidate for re-verification or archiving, especially records tied to open deals.
Step 4: Score and report
Turn your findings into a simple scorecard: one score per dimension, per dataset, with the worst offenders listed. Share it with the people who own each system. Visible problems get fixed.
The Most Common Sources of Bad Data
Fixing records without fixing the source is like bailing water without plugging the leak. These four sources cause most bad data.
Manual entry and broken integrations
People type quickly, copy from memory, and invent shortcuts. Free-text fields invite variation: "Ltd", "Limited", and "LTD" for the same company. The fix is structure: use picklists instead of free text, make critical fields required, and validate formats at entry. Integrations cause the other common failure. A field mapped to the wrong target, a one-way sync, or a failed job can corrupt thousands of records silently. Review integration mappings at least once a quarter, and decide which system is the source of truth for every shared field.
Legacy imports and no ownership
Data imported from an old system or a purchased list carries the old system's habits with it: outdated formats, dead contacts, and duplicates. Never import a list without profiling it first. Clean it in a staging area, match it against existing records, and only then load the survivors. Finally, when nobody owns a dataset, nobody fixes it.
A Realistic Example: How One Sales Team Fixed Its CRM
Consider an illustrative example of a mid-sized B2B sales team. Their CRM held several thousand accounts, and the reporting had stopped making sense. Pipeline totals changed every time someone ran the same report, some accounts appeared twice with different owners, and many contacts had not been verified in over a year.
The team ran the checklist in this article. They mapped their sources and found leads entered through three routes: a website form, a trade-show import, and manual entry by reps. Profiling showed duplicate accounts created whenever the same company arrived through two routes, because nothing matched on domain first. Sorting by last activity revealed many contacts untouched for more than a year.
The fix followed the checklist order. They defined a matching rule on company domain, merged duplicates while keeping the most complete record and its history, and archived stale contacts after a final re-verification attempt. They made the website form validate email format and company domain at entry, replaced free-text industry fields with a picklist, and named one person as CRM data owner with a monthly review. Within a few reporting cycles, pipeline totals stabilised and managers trusted the numbers again. Nothing here required new software, only a checklist, a few rules, and a named owner.
Your Step-by-Step Fix Plan
Once the audit is done, fix in priority order. High-impact, easy fixes come first.
- Stop the bleeding. Add validation rules and required fields at every point of entry, so new bad data stops arriving while you clean the old data.
- Merge duplicates. Start with records tied to open deals and active customers. Always keep the full history on the surviving record.
- Fill the gaps. Enrich missing critical fields for high-value records first, using verified sources. Do not guess values.
- Reconcile systems. Pick a source of truth for every shared field, fix the mismatches, and repair the mapping that caused them.
- Refresh stale records. Re-verify contacts and accounts that matter, and archive the rest.
- Document the rules. Write down every standard you applied, so the next person follows the same rules.
Keeping Data Clean: Habits That Make the Checklist Stick
A one-time cleanup fades within months. Run the full checklist at least once a quarter, and run the fast checks, duplicate rate, field fill rate, and stale-record count, every month. Validate at entry, so every form, import, and integration checks formats, required fields, and duplicates before a record is saved. Give every dataset a named owner. Set thresholds for each dimension and alert the owner when a score drops. Train the team, because a rep who understands that a duplicate account splits their own deal history will enter data more carefully. Finally, review AI inputs separately: before any dataset feeds an AI model, run the checklist on it alone, because AI amplifies whatever it is given, including errors.
For the governance questions that sit alongside quality, see our article on AI data residency and in-country AI processing in 2026.
Clean Data and AI: Why Quality Comes Before Automation
Every AI initiative in 2026 rests on the same foundation: the data it learns from. When that foundation is solid, AI delivers faster analysis and recommendations managers can trust. When it is not, AI delivers confident answers built on bad inputs, at a speed no human review can keep up with. That is why data quality work should come before, not after, an AI rollout. Before you connect a dataset to an AI tool, run the six-dimension checklist on it, fix what fails, and record the baseline.
If you need help building that foundation, our data engineering and analytics services cover data audits, pipeline design, and analytics setup, so your dashboards, reports, and AI tools all run on data you can trust.
Frequently Asked Questions
What is a data quality checklist?
A data quality checklist is a structured list of checks that tests your data against the six dimensions of quality: accuracy, completeness, consistency, timeliness, validity, and uniqueness. It turns a vague concern into a repeatable audit you can run on a schedule and score over time.
How often should a business audit its data?
Run the full checklist at least once a quarter, and run the fast checks, duplicate rate, field fill rates, and stale-record counts, every month. Businesses with fast-changing contact data or heavy AI use should check more often.
What are the six dimensions of data quality?
The six dimensions are accuracy, completeness, consistency, timeliness, validity, and uniqueness. Accuracy asks whether a value matches reality, completeness asks whether required fields are filled, consistency asks whether systems agree, timeliness asks whether data is current, validity asks whether values follow the rules, and uniqueness asks whether each entity appears once.
What causes most bad data in a business?
The most common causes are manual entry errors, broken integrations, uncleaned legacy imports, and a lack of ownership. Manual entry creates typos, broken integrations create mismatches between systems, legacy imports bring old problems into new systems, and no ownership means nobody fixes any of it.
Can AI fix bad data automatically?
AI can help a great deal with detection and cleanup, including fuzzy duplicate matching and standardising formats at scale. But AI cannot decide what is true on its own, and it cannot fix a broken process. The best results come from combining AI tools with clear rules, human review of uncertain matches, and validation at entry.
How do I start fixing data quality with a small team?
Start with one dataset that drives real decisions, usually your CRM or customer list. Run the audit steps in this checklist, fix duplicates and stale records first, add validation at entry, and name one owner.
Bad data is not a one-time problem. It is routine maintenance, like servicing equipment before it breaks down. Work through this data quality checklist once, fix what it finds, and run it again on a schedule. Your dashboards will get steadier, your AI tools will get smarter, and your decisions will rest on facts you can defend. If you would like expert help auditing your data and building a lasting quality routine, contact Skills Technology Solutions.
Related Articles
Keep reading — more guides you may find useful.
Related Services
Services connected to this guide.
