What is data quality?

At a glance

• Data quality is measured against its intended purpose: the same data may be sufficient for monthly reporting but too imprecise for a daily forecast.
• Six core dimensions are considered standard: accuracy, completeness, consistency, timeliness, validity, and uniqueness.
• Data quality can be measured using metrics such as error rate, duplicate rate, or the completion rate of mandatory fields.
• Data errors are least costly when detected during data capture or integration, rather than later in a management report.
• AI amplifies the effect: AI-powered analyses and forecasts only produce useful results when based on validated data.

Data quality definition

Data quality refers to the degree to which data is suitable, accurate, and reliable for a specific purpose. It is assessed based on various quality characteristics such as accuracy, completeness, timeliness, consistency, and uniqueness. High data quality ensures that data can be processed and analyzed reliably and used as a basis for well-founded decisions.

Data quality is not merely a technical detail; it is a key prerequisite for business success. Reliable, up-to-date, and complete data enables companies to generate sound insights and build reliable automated processes. In an era of business intelligence, artificial intelligence (AI), and data-driven decision-making, the quality of the underlying data is critical.

A high standard of data quality means that data must not only be technically correct but also understandable, relevant, and usable in the right context. Ideally, data should support processes rather than hinder them. Companies that recognize this invest systematically in capturing, validating, cleansing, and continuously monitoring their data.

What does data quality mean for a company?

Data quality in a company means that data is fit for its specific purpose: for month-end closing, sales planning, or an AI project. A dataset is never inherently good or bad; its quality always depends on the requirements of its users. In the literature, this principle is known as “fitness for use.”

Data quality goes beyond technical correctness. Data must also be consistently defined, understandable, and usable in the right context. A revenue figure is of little use if sales and controlling interpret it differently – for example, as gross revenue versus net revenue.

In business intelligence and controlling, data quality determines whether reports inspire confidence or trigger discussions about the numbers. Data quality is therefore not a one-time project but an ongoing responsibility shared by business functions and IT.

How does data quality differ from related terms?

Data quality is the measurable outcome that data integrity and data governance make possible.

Why is data quality so important?

Companies with high data quality not only work more efficiently but can also respond more quickly to market changes because they have access to reliable data.

Poor data quality affects:

What are the criteria and dimensions of data quality?

Data quality is generally measured across six core dimensions: accuracy, completeness, consistency, timeliness, validity, and uniqueness. Each dimension can be assessed using its own verification question and set of rules.

What other criteria are relevant?

In addition to the core dimensions, other criteria are important, particularly for use by business functions. The ISO/IEC 25012 standard defines a total of 15 data quality characteristics.

• Relevance: The data answers the questions for which it was collected.
• Understandability: Metrics and fields are clearly defined and documented.
• Provenance transparency: Every figure can be traced back to its source (data lineage).
• Accessibility: Authorized users can access the data on time and without unnecessary effort.

What are the characteristics of good data quality in everyday use?

Good data quality is evident when people no longer debate the numbers themselves but focus on what they mean. Typical characteristics include:

• Reports from different departments show the same values for the same metric.
• Controllers no longer manually correct figures in Excel before publishing them.
• Anomalies in reports have business-related causes, not technical ones.
• Business functions know who is responsible for their data and report errors at the source.

How do you assess and measure data quality?

Data quality is assessed by companies defining validation rules for each dimension, measuring the actual state as a metric, and comparing it against a target value. Without measurement, data quality remains a guess: according to Gartner, 59 percent of organizations do not measure their data quality.

What steps does an assessment follow?

• Determine critical data: Which data feeds into management metrics, planning, or AI applications? That is where the review begins.
• Define validation rules: The business department and IT define, for each dimension, what "correct" means, such as mandatory fields, value ranges, or reconciliation rules between systems.
• Measure the actual state: Data profiling reveals fill rates, value distributions, duplicates, and format errors.
• Set thresholds: A target value is defined for each metric, for example at least 98 percent filled mandatory fields.
• Monitor continuously: The metrics are checked with every data load and made visible in a data quality report.

Which metrics are suitable for measurement?

The thresholds depend on the purpose. For consolidated financial statements, a reconciliation difference of zero is mandatory, while for a trend analysis in marketing lower accuracy is often sufficient.

What impact does data quality have on business results?

Poor data quality costs money, time, and trust. A Gartner study from 2020 puts the cost of poor data quality at an average of at least 12.9 million US dollars per organization per year (Gartner). The figure is based on estimates from large companies and should be understood as an order of magnitude.

The costs rise the later an error is detected. The so-called 1-10-100 rule describes this effect: an error that costs 1 euro at capture costs about 10 euros to clean up later and about 100 euros as a wrong decision. Checks at the source and during data integration are therefore the most economical.

How can data quality be improved?

Data quality improves permanently only when organization and technology work together: clear responsibilities, uniform definitions, and automatic checks in data integration. A one-time cleanup removes symptoms, but not the causes.

Organizational measures

• Name a data owner for each data area, for example sales for customer data and purchasing for supplier data.
• Define metrics in a shared glossary so that "revenue" means the same in all reports.
• Set rules for data capture and train employees in them.
• Have errors fixed at the source instead of overwriting them in the report.

Technical measures

• Validate entries in the source system, for example through mandatory fields and selection lists.
• Build validation rules into ETL processes so that faulty records are caught before loading.
• Consolidate data in a central data basis that serves as a single source of truth.
• Detect and merge duplicates using matching rules.
• Measure and report data quality metrics with every load.

How do you improve data quality with BI tools?

BI tools improve data quality because they make errors visible where business users notice them: in reports, variance analyses, and drill-downs. A controller spots an implausible monthly value faster than any generic validation rule.

Which errors occur frequently and how can they be avoided?

The most common error is treating data quality as a one-time IT project. Without ongoing checks and clear responsibilities, old errors quickly return.

What role does data quality play for artificial intelligence?

Data quality and artificial intelligence are connected in two directions: AI needs verified data, and AI helps find data errors. Both sides gain importance as AI is adopted in controlling.

Why does AI need high data quality?

AI models adopt the errors in their data and pass them on as seemingly well-founded results. The principle is called "garbage in, garbage out". An AI forecast based on incomplete actual data looks plausible and is nevertheless wrong.

With language models, another level is added: if a large language model accesses documents via retrieval-augmented generation (RAG), the timeliness and clarity of texts also matter. Outdated guidelines or duplicate versions lead to contradictory answers.

Lawmakers also impose requirements: Article 10 of the European AI Regulation (EU AI Act) requires that, for high-risk AI systems, the training, validation and testing data be relevant, sufficiently representative, and to the best extent possible free of errors and complete in view of the intended purpose (Art. 10 AI Act).

How does AI improve data quality?

AI recognizes patterns that fixed validation rules overlook. Typical fields of application are:

• Detecting anomalies and implausible values in large data volumes
• Tracking down duplicates with differing spelling
• Classifying unstructured information, for example from conversation records
• Suggesting mappings and cleansing steps when connecting new sources

The AI data integration from Bissantz automatically detects format breaks, typos and implausible values, suggests cleansing and reduces manual review loops. The functional assessment of the suggestions remains with humans, supported by Explainable AI.

Which trends shape the topic of data quality?

The most important trends move data quality forward into data integration and make it a prerequisite for AI. Instead of looking for errors in the finished report, companies increasingly check data automatically and continuously.

Practical example: data quality in sales controlling

A mid-sized manufacturer reports its monthly revenue from the ERP system and from the BI system. The values regularly differ, and before every closing the controlling team spends days reconciling in Excel. Management doubts the figures of the sales planning.

The analysis reveals three causes: after connecting the CRM system, customers exist twice, new items are not assigned to any product group, and credit notes are posted in the systems at different points in time.

The result: causes are corrected in the upstream system instead of overwritten in the report. Controlling comments on remaining anomalies directly in DeltaMaster, and management again discusses measures instead of figures.

Application examples: where data quality is decisive

Data quality affects almost all data-driven processes in a company. The following examples show typical areas of use in which reliable data determines sound decisions.

Summary

Data quality is the silent foundation of all data-driven activities: only when data is accurate, complete, consistent, and up to date can analyses, reports, and AI applications deliver reliable results. It is not a one-time project but an ongoing process, embedded in data models, ETL processes, and an established data culture. With DeltaMaster and its consulting expertise, Bissantz makes data quality an integral part of corporate management—visible, manageable, and sustainably secured.