Last week, I heard a story about a VP who greenlit a $200,000 marketing campaign based on what looked like rock-solid customer data. Three weeks in, they noticed something odd: conversions were 60% lower than projected.
Turns out, the report was counting duplicate customer IDs. The "new customer surge" they thought they had? Just the same people showing up twice in the spreadsheet.
The Real Problem
The campaign wasn't the problem. The data was. And here's the scary part: the numbers looked perfectly fine. Clean columns. Nice charts. Professional dashboard. No red flags.
Bad data doesn't announce itself.
It sits there, quietly, looking legitimate — until you make a decision that costs real money, wastes real time, or embarrasses you in front of your team.
So how do you avoid this? You run checks. Simple ones. Before you trust any data, before you build a strategy around it, and definitely before you spend a dollar on it.
Here are the 10 checks that will save you from making decisions on bad data.
Wake-Up Call
Bad data costs organizations an average of $15 million annually, yet 47% of newly created data records have at least one critical error.
Why Most Data Problems Are Silent
The biggest misconception about data quality is that bad data will look bad. It won't.
A spreadsheet with missing values doesn't crash. A dashboard with incorrect dates doesn't throw an error. A report with duplicates doesn't come with a warning label.
Everything looks normal until someone asks a question the data can't answer correctly.
That's when you find out your "top-performing region" was actually just reporting data twice. That's when you realize your "customer growth rate" included test accounts. That's when you discover the pricing column has been pulling from last quarter's data for three months.
Data quality issues hide in plain sight. And the people who spot them early aren't doing anything magical. They're just running the same checks, every time, before they trust what they're looking at.
The Data Quality Mindset
Before we dive in, remember this: Your job isn't to trust data. Your job is to verify it. Every dataset is guilty until proven innocent. This mindset alone will save you from more bad decisions than any tool or technique.
The 10 Data Quality Checks
1. Check for Missing Values
What it means: Look for blank cells, nulls, or placeholders like "N/A" or "TBD."
Why it matters: Missing data skews averages, throws off counts, and makes analysis unreliable. If 30% of your revenue column is blank, your revenue total is wrong.
Real Example: You're analyzing customer support response times. You filter for "average response time under 2 hours" and feel great about performance. But 40% of tickets don't have a response time logged at all. Those are probably the ones that took 6 hours. Your "fast response" metric just became meaningless.
2. Check for Uniqueness
What it means: Make sure things that should be unique actually are — customer IDs, order numbers, email addresses.
Why it matters: Duplicates inflate your counts and make metrics like "total customers" or "total sales" completely unreliable.
Real Example: You pull a list of 5,000 customers who bought in Q4. Looks great — until you realize 1,200 appear twice because someone merged two systems without deduplicating. Your actual customer count? 3,800. That's a 24% difference, and every decision you make on "5,000 customers" is wrong.
3. Check for Broken Relationships
What it means: Make sure data that's supposed to connect actually does. Orders should have customers. Invoices should have line items. Transactions should have accounts.
Why it matters: Broken links create orphan records that either get ignored (undercounting) or cause confusion (mismatching).
Example: You're reviewing sales by rep. You see 300 orders with no rep assigned. Are those self-service? Web orders? A data entry problem? If you don't know, your "top performer" ranking is useless — and your commission calculations might be wrong.
Quick Tip
When you find broken relationships, don't just fix the symptoms. Ask why they broke in the first place. Often, it reveals a process problem that will keep creating bad data until you address the root cause.
4. Check for Accepted Values
What it means: Fields that should have a limited set of options (like status: "active," "inactive," "pending") shouldn't have random values like "activ" or "Pending" or "TBD."
Why it matters: Typos and inconsistencies break your filters and counts. If you filter for "active" but some records say "Active" or "ACTIVE," you'll miss data.
Real Example: You're counting how many projects are "Complete." The system shows 120. But someone also entered "Completed," "Done," and "Finished" in other rows. Your real count? 180. That's a 50% undercount — and it makes your team look slower than they actually are.
5. Check Business Logic
What it means: Does the data make sense in the real world? Can someone have a purchase date before their account was created? Can a refund be larger than the original sale?
Why it matters: Data can be "technically correct" but logically impossible. These contradictions expose errors earlier in the process.
Real Example: You see a customer who signed up on January 15th but made a purchase on January 10th. That's impossible. Either the signup date is wrong, the purchase date is wrong, or this record was merged from another system incorrectly. Until you fix it, any "time to first purchase" analysis is broken.
6. Check for Reasonable Ranges
What it means: Numbers should fall within expected boundaries. Ages between 0–120. Prices above $0. Percentages between 0–100%.
Why it matters: Outliers that break reality signal data entry errors or system glitches. A single bad value can ruin an average or total.
Example: You're calculating average order value. It's coming out to $8,400, which feels high. You dig in and find one order for $500,000. Turns out someone added an extra zero during data entry. That one mistake just made all your AOV analysis useless.
The Single Bad Value Problem
One outlier can destroy an average. One negative number where there should be none can throw off a total. Always check the extremes, not just the middle.
7. Check Data Types
What it means: Make sure numbers are stored as numbers, dates as dates, and text as text. A price stored as text can't be summed. A date stored as text can't be sorted correctly.
Why it matters: The wrong data type breaks calculations, sorting, and filtering. You'll get weird results and not know why.
Technical Trap
Example: You try to calculate total revenue and get an error. Turns out the revenue column has values like "$1,200" (with the dollar sign). That's text, not a number. Your spreadsheet can't add it. You'll need to clean every single row before your totals work.
8. Check Freshness
What it means: Is the data current? When was it last updated? Are you looking at today's numbers or last month's?
Why it matters: Stale data leads to decisions based on outdated reality. The world moved on, but your data didn't.
The Stale Data Problem
27% of business decisions are made on data that's more than a week old. In fast-moving industries, that's ancient history.
Example: You pull a dashboard showing "current inventory levels." The dashboard hasn't refreshed in three days. The warehouse actually received a shipment yesterday. Now you're about to over-order and tie up cash in inventory you don't need.
9. Check Time Consistency
What it means: Make sure time-based data is measured consistently. Are all dates in the same timezone? Are all timestamps using the same format?
Why it matters: Inconsistent time tracking creates false patterns and makes trend analysis unreliable.
Example: You're tracking daily sign-ups across multiple regions. Your chart shows a massive spike every day at 5 PM. Turns out the European data is in UTC but the US data is in EST. The "spike" is just a timezone mismatch creating a phantom trend.
Time Zone Trap
Time inconsistencies are one of the most common and hardest-to-spot data quality issues. Always verify that timestamps across different systems use the same timezone — or clearly document when they don't.
10. Check for Sudden Changes
What it means: Look for spikes, drops, or shifts that don't match reality. Did your traffic triple overnight? Did your conversion rate suddenly drop to zero?
Why it matters: Dramatic changes are usually data errors, not business changes. Trusting them leads to panic or false confidence.
Example: Your sales dashboard shows a 90% drop in revenue yesterday. But then you check — the payment processor had an outage, so yesterday's sales didn't sync yet. The drop isn't real. If you'd acted on that number, you might have made a very bad call.
Real-World Win
A retail client running these 10 checks discovered that their "declining sales" were actually a reporting bug that double-counted returns. Once fixed, they realized sales were actually up 12% — which completely changed their inventory strategy and saved them from cutting marketing spend at the worst possible time.
Data Trust Is Earned, Not Assumed
Here's what most people miss about data quality: good-looking data isn't the same as trustworthy data.
A dashboard can have beautiful charts and still be built on garbage. A spreadsheet can be perfectly formatted and completely wrong. A report can have executive-level polish and lead you to a terrible decision.
These 10 checks aren't about being paranoid. They're about being smart.
Your Data Quality Checklist
- **Missing values:** Are there gaps in critical columns? - **Uniqueness:** Are there duplicates where there shouldn't be? - **Relationships:** Do the connections between data make sense? - **Valid values:** Are all entries using accepted formats and options? - **Business logic:** Does the data reflect reality? - **Ranges:** Are the values within reasonable boundaries? - **Data types:** Is everything stored correctly? - **Freshness:** Is this data current and up-to-date? - **Time consistency:** Are timestamps and timezones aligned? - **Sudden changes:** Do dramatic shifts match real events?The people who run these checks don't have more data than you. They just trust it less until it proves itself.
Automation makes bad data faster. It doesn't make it better. The next time you're about to make a decision based on data, pause. Run these checks. Ask the uncomfortable questions. Look for the things that don't add up.
Good decisions deserve good data. Don't settle for anything less.
Need Help With Your Data Quality?
We help businesses implement data quality checks that catch errors before they cost you money — from simple validation to enterprise-wide monitoring.
Schedule a free data audit →