How do you prove to your boss that your data is trustworthy enough to make million-dollar decisions? That is the core problem that 'Data Quality Metrics, Monitoring, and Remediation' solves, and it is a central topic on the DA0-002 exam. You will learn to measure data quality like a scientist, watch for problems like a security guard, and fix errors like a mechanic — all essential skills for any data analyst.
Jump to a section
A simple way to picture Data Quality Metrics, Monitoring, and Remediation
A busy restaurant kitchen on a Friday night. The head chef is not just cooking — they are watching every plate that goes out.
That plate of pasta? The chef checks the portion size, makes sure the sauce is the right temperature, and confirms the garnish is present. These are data quality metrics: specific, measurable checks (like completeness, timeliness, and accuracy) that tell you if your data is good enough to use.
Now, the chef does not just check one plate. They have a system. A thermometer probe is used every hour to log fridge temperatures. A checklist is ticked for every order. This is monitoring: continuously measuring those metrics over time to spot when things start to go wrong. When the thermometer shows the fridge is two degrees too warm, that is a monitored metric triggering an alert.
Finally, what happens when the chef finds a plate with undercooked chicken? The plate does not go to the customer. It gets sent back to the line cook, the cooking time is increased, and the supplier of that chicken batch is flagged. That is remediation: finding the root cause of a data quality problem, fixing the bad data, and changing the process so it does not happen again. Without these three steps — metrics, monitoring, remediation — the kitchen would serve bad food, and the data kitchen would serve bad decisions.
Data quality is about how fit your data is for its intended purpose. Bad data leads to bad decisions, unhappy customers, and failed projects. To manage data quality, you need three things: metrics to measure it, monitoring to track it over time, and remediation to fix it when it breaks. This section explains each piece from the ground up.
First, let us define data quality dimensions. These are the specific categories you use to measure quality. The DA0-002 exam expects you to know these six key dimensions:
Accuracy: Is the data correct? A customer's address stored as '123 Main St' when they actually live at '124 Main St' is inaccurate.
Completeness: Are all the required data values present? If a sales record has a missing 'Region' field, that record is incomplete.
Consistency: Does the data conflict with itself? If a customer's age is '35' in one system and '40' in another, you have an inconsistency.
Timeliness: Is the data available when needed? A stock price from yesterday is useless for a day trader — it is not timely.
Uniqueness (or Deduplication): Are there duplicate records? Two identical customer profiles for the same person waste resources and cause confusion.
Validity: Does the data conform to the expected format? A phone number stored as 'abc-def-ghij' is invalid because phone numbers should be numeric.
Now, how do you measure these? You design metrics. A metric is a specific, quantifiable measurement. For example, you could calculate a 'completeness score' for your customer database by dividing the number of records with a complete address by the total number of records. You might aim for 99% completeness. For accuracy, you could run a random sample of 100 records and manually check them against source documents, then calculate the percentage that are correct.
Monitoring is the process of repeatedly collecting these metrics over time. You do not just measure once; you set up a dashboard or a recurring report. For example, you could run a completeness check on your customer database every Sunday night. If the completeness score drops below 95%, the system sends an email alert to the data team. Monitoring helps you spot trends — maybe accuracy starts dropping after a new data entry person is hired, so you can intervene early.
Finally, remediation is the process of fixing problems when they are detected. Remediation has two parts: corrective action and preventive action. Corrective action fixes the immediate problem — for example, updating a single wrong address in the database. Preventive action changes the process so the problem does not recur — for instance, adding a validation rule that prevents anyone from entering a non-numeric phone number. A good remediation process records what went wrong, what was done to fix it, and what was done to prevent it. This documentation is often called a 'data quality issue log'.
In the old days, data quality was ignored or fixed reactively after a disaster — like a customer getting a bill for someone else's purchase. Modern data management uses these three processes proactively. You define your metrics before you start a project. You monitor them constantly. And you have a standard operating procedure for remediation. This structured approach is what the DA0-002 exam is testing: can you identify which metric to use, when monitoring is needed, and what the correct remediation step is for a given scenario.
Define Data Quality Metrics
Before you can manage quality, you must decide what to measure. For each important data field, choose one or more dimensions (e.g., accuracy, completeness) and define a specific metric. For example, 'the percentage of customer records with a valid email address' is a completeness and validity metric. This step is critical because it sets the benchmark for all future monitoring.
Establish Baseline Values
Run the metrics for the first time to see the current state of your data. Record these baseline values. For instance, you might find that 95% of email addresses are valid today. The baseline gives you a starting point to measure improvement or decline.
Set Up Monitoring and Thresholds
Decide how often you will collect each metric (daily, weekly, monthly) and what threshold triggers an alert. For example, if email validity drops below 90%, send an email to the data team. Monitoring without thresholds is pointless because you will not know when to act.
Detect and Log the Issue
When a monitored metric breaches its threshold, create a data quality issue log entry. Record the date, the specific metric that triggered the alert, the data involved, and the initial description of the problem. This step ensures the issue is documented and can be tracked.
Perform Corrective Action
Fix the immediate data errors. For example, update the invalid email addresses in the database using a script or manual correction. This stops the immediate impact on business operations, such as failed email campaigns.
Perform Preventive Action
Identify the root cause of the error and change the process to prevent it from recurring. This could involve adding a validation rule at the point of data entry, updating a training manual, or modifying a data import script. Without this step, the same error will likely reappear.
Review and Improve the Process
Regularly (e.g., quarterly) review the data quality issue log, the effectiveness of preventive actions, and whether the current metrics and thresholds are still appropriate. This continuous improvement loop keeps your data quality management effective as the business evolves.
Meet Priya, a junior data analyst at a mid-sized online retailer. Her boss has just received a complaint: a customer was sent a children's book when they ordered a laptop. The cause? The warehouse picked the wrong product because the product SKU (a unique stock-keeping unit identifier) was incorrectly recorded in the inventory system. Priya's boss asks her to fix this and make sure it never happens again.
Priya starts with metrics. She needs to know how bad the problem is. She runs a query on the inventory database and calculates the percentage of products whose SKU matches the master product list. This is an accuracy metric. She finds that 2% of SKUs are mismatched. She also checks the completeness of the 'product description' field, finding that 5% of descriptions are blank. She documents these as baseline metrics.
Next, she sets up monitoring. She creates a weekly report that runs automatically on Monday morning. The report recalculates the SKU accuracy and product description completeness metrics. She sets a threshold: if SKU accuracy falls below 99%, an alert is sent to the inventory team. She also builds a simple dashboard in the company's analytics tool showing a line chart of these metrics over the past month. This way, the team can see if the problem is getting better or worse.
Now, remediation. For the single wrong SKU that caused the laptop mix-up, Priya manually corrects it in the database. That is corrective action. For preventive action, she works with the IT team to add a 'picklist' in the data entry form — instead of typing a SKU manually, users must select it from a valid list. This prevents typographical errors. She also creates a formal 'Data Quality Issue Log' in a shared spreadsheet. The log includes: date of issue, description, root cause, corrective action taken, preventive action implemented, and the person responsible.
Priya presents her findings to the team. She shows the dashboard, explains the new validation rule, and schedules a monthly review meeting. Six months later, SKU accuracy is at 99.8%, and no more wrong product shipment complaints have come in. Priya's combination of metrics, monitoring, and remediation saved the company money and customer trust.
The DA0-002 exam dedicates a significant portion of Domain 5.3 to data quality metrics, monitoring, and remediation. Expect multiple-choice questions that test your ability to match a data quality problem with the correct dimension, metric, or remediation step. You will not be asked to write SQL or configure a monitoring tool, but you must be able to read a scenario and identify the right concept.
Here is what the exam tests specifically:
Identifying the six data quality dimensions: accuracy, completeness, consistency, timeliness, uniqueness, and validity. A common question gives you a description of a data problem and asks which dimension is violated. For example, 'A customer record has two different phone numbers for the same person' — the answer is consistency or uniqueness, depending on context.
Calculating a simple metric: You might be given a sample of 500 records, told that 480 have a complete 'email' field, and asked to calculate the completeness percentage. The answer is 96% (480/500 * 100).
Selecting the appropriate remediation action: A scenario will describe a data quality issue found during monitoring. The question will ask whether the correct first step is corrective action (fix the bad data) or preventive action (change the process). The trap is that beginners often jump to prevention before fixing the immediate problem. The exam pattern: corrective action first, then preventive.
Interpreting a monitoring dashboard: A mock dashboard shows a metric trending downward for three months. The correct answer is 'the data quality is deteriorating' and 'remediation is needed'. The trap: ignoring the trend and focusing on the current value alone.
Understanding the data quality issue log: The exam may ask what information should be recorded when a data quality issue is found. The correct answer includes: date, description, root cause, corrective action, preventive action, and owner.
Key definitions to memorise:
Data quality metric: A quantifiable measurement of a data quality dimension.
Data quality monitoring: The ongoing process of collecting and reviewing data quality metrics.
Data quality remediation: The process of fixing data errors and preventing their recurrence.
Corrective action: Fixing the immediate data error.
Preventive action: Changing the process to stop the error from happening again.
Common traps on the exam:
Confusing 'accuracy' with 'validity'. Accuracy means the data reflects reality (correct value). Validity means it matches the defined format (correct type). A phone number could be valid (all numbers) but inaccurate (wrong phone number).
Mistaking 'timeliness' for 'completeness'. A report that arrives two weeks late is not timely, even if it contains all the required fields.
Assuming monitoring always requires expensive software. The exam acknowledges that monitoring can be done with simple scripts and scheduled reports.
Thinking remediation ends with fixing the data. The exam expects you to always include a preventive step.
Practise reading scenarios carefully. Look for keywords: 'wrong value' suggests accuracy; 'missing field' suggests completeness; 'duplicate record' suggests uniqueness; 'format error' suggests validity; 'conflicting values' suggests consistency; 'delayed report' suggests timeliness.
Data quality is measured across six dimensions: accuracy, completeness, consistency, timeliness, uniqueness, and validity.
A data quality metric is a specific, quantifiable measurement of one dimension, such as the percentage of complete records in a dataset.
Monitoring involves collecting metrics repeatedly over time and setting thresholds to trigger alerts when quality drops.
Remediation has two required steps: corrective action to fix the immediate data error, and preventive action to stop the error from happening again.
The importance of a quality dimension depends on the business use case — a medical record needs high accuracy; a mailing list needs high completeness.
A data quality issue log should record the date, description, root cause, corrective action, preventive action, and responsible person for every problem found.
Monitoring without thresholds is just data collection — you need defined alerts to know when to act.
Data quality management is a continuous cycle: define metrics, monitor them, fix issues, and improve processes, then repeat.
Confusing accuracy (correct value) with validity (correct format) is a common exam trap.
Preventive actions include adding validation rules, improving data entry forms, and training staff.
These come up on the exam all the time. Here's how to tell them apart.
Data Quality Dimension
A category of quality, e.g., accuracy, completeness.
Describes what aspect of data you care about.
Does not include a number or measurement.
Data Quality Metric
A specific, quantifiable measurement, e.g., 95% complete.
Implements the dimension with a concrete value and formula.
Includes a numerical result that can be compared over time.
Corrective Action
Fixes the immediate data error (e.g., updates a wrong address).
Reactive — done after the problem is detected.
Does not address the root cause of the error.
Preventive Action
Changes the process to stop the error from recurring (e.g., adds a validation rule).
Proactive — done after corrective action to prevent future issues.
Addresses the root cause of the error.
Data Quality Monitoring
An ongoing, continuous process of measuring metrics.
Typically automated with dashboards and alerts.
Designed to catch problems quickly in near real-time.
Data Quality Audit
A periodic, formal review of data quality, often by an external team.
Usually done manually or semi-manually on a scheduled basis (e.g., annually).
Provides a comprehensive, deep-dive assessment, not day-to-day status.
Accuracy
Measures if the data value is correct in the real world.
Example: A customer's age stored as '35' when they are actually 35.
Requires a trusted source to verify correctness.
Validity
Measures if the data conforms to a defined format or type.
Example: A phone number stored as 'abc-def-ghij' is invalid because it uses letters.
Can be checked with rules (e.g., regex patterns) without a real-world source.
Mistake
Data quality is only about cleaning bad data after it is already in the system.
Correct
Data quality management is proactive, not just reactive. It includes defining metrics before you collect data, monitoring continuously, and implementing preventive controls to stop errors from being created in the first place.
People often think of data quality as 'data cleaning' — a one-time fix. In reality, modern data governance emphasises prevention and continuous monitoring.
Mistake
All data quality dimensions are equally important for every dataset.
Correct
The importance of each dimension depends on how the data will be used. For a medical diagnosis system, accuracy is critical; for a marketing mailing list, completeness of address might be more important than uniqueness.
Beginners learn a list of dimensions and assume they always need to optimise all of them. The exam tests your ability to prioritise based on context.
Mistake
A single data quality metric, like 99% completeness, guarantees the data is usable.
Correct
A single metric only measures one dimension. Data could be 99% complete but 90% inaccurate. You need a combination of metrics across multiple dimensions to get a true picture of fitness for use.
It is tempting to focus on just one number. But the exam expects you to recognise that data quality is multi-dimensional.
Mistake
Once you fix a data quality issue, you can stop monitoring that metric.
Correct
Monitoring must be ongoing. Processes change, new errors can be introduced, and old problems can recur. You only stop monitoring if the metric is no longer relevant to business needs.
People treat a fix as permanent. The exam emphasises continuous improvement and monitoring as a loop, not a one-time project.
Mistake
Remediation only means correcting the bad data values.
Correct
Full remediation has two parts: corrective action (fixing the data) and preventive action (changing the process to prevent recurrence). The exam specifically tests that you include both.
When you are under pressure, it is easy to just fix the visible symptom. The exam wants you to show you understand root cause analysis and process improvement.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
A dimension is a category of quality, like accuracy or completeness. A metric is a specific way to measure that dimension, such as 'the percentage of records with a correct phone number'. Think of dimensions as the 'what' and metrics as the 'how much'.
It depends on how quickly the data changes and how critical it is. For a real-time customer-facing system, you might monitor every hour. For a monthly sales report, weekly or monthly checks are sufficient. The key is to set a frequency that catches problems before they cause damage.
The first step is corrective action: fix the bad data to stop immediate harm. For example, correct a wrong customer address so the shipment goes to the right place. Only then should you investigate the root cause and implement preventive action.
No. In practice, data quality is about being 'fit for purpose' — good enough for the intended use. Trying to achieve 100% perfection in all dimensions is usually too expensive and time-consuming. The goal is to meet the business requirements for accuracy, completeness, etc.
You can start with simple tools like spreadsheets and scheduled SQL queries. Many organisations use dedicated data quality software, but the DA0-002 exam focuses on concepts, not specific tools. The important thing is the process: define metrics, collect them regularly, and act on the results.
It is a document, often a spreadsheet, where you record every data quality problem you find. It typically includes the date, description, root cause, corrective action taken, preventive action taken, and the person responsible. It helps track issues over time and prove that you are managing quality.
You've finished Data Quality Metrics, Monitoring, and Remediation. Continue through the DA0-002 study guide to build a complete picture of the exam.
Done with this chapter?