Improving Data Quality: Criteria, Measurement and 7 Measures

What Is Data Quality?

Data quality describes how fit data is for its purpose. A customer address is of high quality if the invoice arrives; an item number, if it means the same part in the warehouse, purchasing and sales. Perfect data does not exist – what matters is whether it is good enough for the decisions and processes built on it.

This is exactly where mid-sized companies often struggle. Customer data sits twice in the CRM and the ERP, price lists in Excel, knowledge about special cases in people's heads. Each system works on its own – together they produce contradictory numbers. Data quality is therefore less a question of software than of whether data is captured, maintained and connected properly once. Our article on the data management system describes the role central data management plays in this.

The Six Criteria of Data Quality

To keep "good data" from remaining a gut feeling, a set of six criteria has proven itself. The literature usually calls them data quality dimensions; the UK government's data quality framework, among others, uses them [1]:

In practice, the criteria are linked: a duplicate violates uniqueness and almost always leads to contradictory values as well. To get started, it is enough to measure the three or four criteria that cause the greatest damage in your most important processes.

What Poor Data Quality Costs

The costs of bad data are rarely visible because they are spread across many workplaces: queries, duplicate maintenance, manual corrections, misaddressed invoices, complaints. Gartner puts the average cost of poor data quality at 12.9 million US dollars per year per organization [2]. It is an average that cannot be transferred one to one to a mid-sized company – but it shows the scale of the problem.

There is also a legal obligation: the GDPR requires personal data to be accurate and, where necessary, kept up to date; companies must take every reasonable step to ensure that inaccurate data is erased or rectified without delay [3]. For customer, applicant and employee data, data quality is therefore not a nice-to-have but part of compliance. And anyone who wants to use AI pays for bad data twice – more on that below.

Where Bad Data Comes From

The causes are similar in almost every company:

• Manual entry without validation rules: free-text fields in which the same city is spelled three different ways.
• Data silos: the same data is maintained in several systems – in the ERP, in the CRM and alongside them in Excel lists or Access databases, reconciled via copy-paste or export.
• Missing ownership: nobody is responsible, so everyone maintains a little – or nobody does.
• Aging: people move, companies rename, contacts change. Without maintenance, data becomes outdated on its own.
• Migrations and legacy systems: when moving to a new system, old clutter comes along unless someone cleans up first. Our article on data migration from legacy systems shows how to avoid this.

Measuring Data Quality: Metrics Instead of Gut Feeling

Before you improve data quality, you should know where you stand. The first step is data profiling: an automated analysis that shows for each field how many values are missing, which formats occur and where outliers lie. In our experience, even a simple run over customer master data reveals surprises.

Profiling turns into metrics that you collect regularly – such as the completeness, duplicate and error rate per data type. What matters is less the perfection of the measurement than its regularity: a monthly traffic light for the three most important metrics makes progress visible and setbacks recognizable early. Our article on master data management shows what this looks like for master data.

Improving Data Quality: 7 Measures

1. Inventory: Check the Data Instead of Guessing

Start with the data type that causes the most damage – usually customer or item master data. Measure the six criteria, prioritize by business impact and decide which values you want to reach in three months.

2. Clean Up: Remove Duplicates and Legacy Clutter

Merge duplicates, standardize formats, archive dead records: this basic cleanup is a one-time effort with immediate effect – and the prerequisite for the following measures to work. Our data hygiene audit describes how we approach it.

3. First Time Right: Prevent Errors at Entry

The cheapest correction is the one that is never needed. Required fields, picklists instead of free text, format checks and a duplicate warning on creation ensure that new data is right from the start. Every validation rule at the source saves many times the rework later.

4. Define One Leading Source per Data Type

Each data type needs one system that leads – all others take over from there. This creates the golden record, the one reliable version of a customer or item. Our service data consolidation for master data shows how scattered records come together.

5. Clarify Ownership: Data Owner and Data Steward

Data quality needs people in charge. The data owner is responsible for a data type and decides on rules; the data steward takes care of maintenance and checks day to day. In mid-sized companies, these are not new positions but clearly assigned tasks. Our data governance guide describes the role model.

6. Connect Silos Instead of Copying Data

Every manual transfer between systems is a source of errors. Interfaces that automatically bring data to where it is needed eliminate typos, version chaos and outdated copies – often the fastest lever of all. Our overview of data integration tools shows which tools are suitable.

7. Measure, Train and Adjust Continuously

Data quality is not a project with an end date. Automated checks flag anomalies before they cause damage; regular reports show whether the measures are working. And your team needs to know why clean data matters: a short training that shows the consequences of a typo often works better than any rule.

Our case study from a machinery manufacturer shows how these steps work together: from a data quality audit and a central master data registry to automated checks in the ERP workflows.

How good is your data really? With our data maturity assessment, we measure quality, silos and ownership and deliver a prioritized roadmap. The fastest way is a direct conversation: book a 30-minute intro call.

Which Software Helps With Data Quality?

There is no software that creates data quality at the push of a button. What works is the interplay of systems that prevent errors at the source and tools that check and maintain the existing data. These software groups have proven themselves:

First check what your existing systems can already do: many CRM and ERP systems come with required fields, validation rules and duplicate checks – our article on CRM process automation shows how to use them in sales. Additional software only pays off once it is clear which rule it is supposed to enforce. Otherwise it merely automates the chaos.

Data Quality and AI: Data First, Then AI

AI amplifies what it finds. An assistant that accesses contradictory customer data gives contradictory answers; a forecast based on incomplete sales data delivers false precision. That is why data quality comes at the beginning of every AI project – not at the end. More on this in our article on data management as the foundation for AI projects.

Conversely, AI now also helps with cleaning up: it detects duplicates despite different spellings, suggests matches and finds outliers in large datasets. What remains decisive is the person who sets the rules and checks the suggestions.

Frequently Asked Questions About Data Quality

What Is Data Quality Management?

Data quality management covers the rules, roles, processes and tools a company uses to secure the quality of its data permanently: measuring, cleaning, preventing errors and assigning responsibility. It is part of data governance. The difference from data quality: data quality describes the state of the data, data quality management the ongoing task of measuring and improving that state.

What Does Data Integrity Mean?

Data integrity means that data remains complete, accurate and consistent throughout its life cycle – even when it is stored, transferred or changed. In databases, integrity rules ensure this, for example unique keys or the requirement that every order refers to an existing customer.

How Long Does It Take to Improve Data Quality?

An inventory with initial metrics takes days, cleaning the most important data type usually a few weeks. The effect becomes noticeable as soon as validation rules apply at entry: from then on, the stock of clean data grows on its own. If there is an acute problem, our data emergency room stabilizes the situation at short notice.

Conclusion

Improving data quality does not mean making all data perfect. It means making the most important data measurable, unique and owned – and preventing errors where they arise. Anyone who starts with one data type, cleans it up, defines a leading source and secures data entry sees the effect within weeks. And lays the foundation for automation and AI along the way.

References

1. (UK Government Data Quality Hub – Meet the data quality dimensions)
2. (Gartner – How to Improve Your Data Quality)
3. (Art. 5 GDPR – Principles relating to processing of personal data)