The following excerpt from Data Quality: The Accuracy Dimension is printed with permission from from Morgan Kaufmann,...
a division of Elsevier. Copyright 2003. Click here to access the complete chapter on Data quality assurance.
Goals of a data quality assurance program
A data quality assurance program is an explicit combination of organization, methodologies, and activities that exist for the purpose of reaching and maintaining high levels of data quality. The term assurance puts it in the same category as other functions corporations are used to funding and maintaining. Quality assurance, quality control, inspection, and audit are terms applied to other activities that exist for the purpose of maintaining some aspect of the corporation's activities or products at a high level of excellence. Data quality assurance should take place alongside these others, with the same expectations.
Just as we demand high quality in our manufactured products, in our financial reports, in our information systems infrastructure, and in other aspects of our business, we should demand it from our data.
The goal of a data quality assurance program is to reach high levels of data accuracy within the critical data stores of the corporation and then keep them there. It must encompass all existing, important databases and, more importantly, be a part of every project that creates new data stores or that migrates, replicates, or integrates existing data stores. It must address not only the accuracy of data when initially collected but accuracy decay, accurate access and transformation of that data, and accurate interpretation of the data for users. Its mission is threefold: improve, prevent, monitor.
Improvement assumes that the current state of data quality is not where you want it to be. Much of the work is to investigate current databases and information processes to find and fix existing problems. This effort alone can take several years for a corporation that has not been investing in data quality assurance.
Prevention means that the group should help development and user departments in building data checkers, better data capture processes, better screen designs, and better policies to prevent data quality problems from being introduced into information systems. The data quality assurance team should engage with projects that build new systems, merge systems, extract data from new applications, and build integration transaction systems over older systems to ensure that good data is not turned into bad data and that the best practices available are used in designing human interfaces.
Monitoring means that changes brought about through data quality assurance activities need to be monitored to determine if they are effective. Monitoring also includes periodic auditing of databases to ensure that new problems are not appearing.
Structure of a data quality assurance program
Creating a data quality assurance program and determining how resources are to be applied needs to be done with careful thought. The first decision is how to organize the group. The activities of the group need to be spelled out. Properly skilled staff members must be assigned. They then need to be equipped with adequate tools and training.
Data Quality Assurance Department
There should be a data quality assurance department. This should be organized so that the members are fully dedicated to the task of improving and maintaining higher levels of data quality. It should not have members who are part-time. Staff members assigned to this function need to become experts in the concepts and tools used to identify and correct quality problems. This will make them a unique discipline within the corporation. Figure 4.1 is a relational chart of the components of a data quality assurance group.
The group needs to have members who are expert data analysts. Analyzing data is an important function of the group. Schooling in database architecture and analytical techniques is a must to get the maximum value from these activities. It should also have staff members who are experienced business analysts. So much of what we call quality deals with user requirements and business interpretation of data that this side of the data cannot be ignored.
The data quality assurance group needs to work with many other people in the corporation. It needs to interact with all of the data management professionals, such as database administrators, data architects, repository owners, application developers, and system designers. They also need to spend a great deal of time with key members of the user community, such as business analysts, managers of departments, and web designers. This means that they need to have excellent working relationships with their customers.
One way to achieve a high level of cooperation is to have an advisory group that meets periodically to help establish priorities, schedules, and interactions with the various groups. This group should have membership from all of the relevant organizations. It should build and maintain an inventory of quality assurance projects that are worth doing, keep this list prioritized, and assign work from it. The advisory group can be very helpful in assessing the impact of quality problems as well as the impact of corrective measures that are subsequently implemented.
Data quality assurance methods
Figure 4.2 shows three components a data quality assurance program can build around. The first component is the quality dimensions that need to be addressed. The second is the methodology for executing activities, and the last is the three ways the group can get involved in activities.
The figure highlights the top line of each component to show where a concentration on data accuracy lies. Data accuracy is clearly the most important dimension of quality. The best way to address accuracy is through an insideout methodology, discussed later in the book. This methodology depends heavily on analysis of data through a process called data profiling. The last part of this book is devoted to explaining data profiling. Improving accuracy can be done through any of the activities shown. However, the one that will return the most benefit is generally the one shown: project services.
Any data quality assurance function needs to address all of the dimensions of quality. The first two, data accuracy and completeness, focus on data stored in corporate databases. The other dimensions focus on the user community and how they interpret and use data.
The methods for addressing data quality vary as shown in Figure 4.3. Both of these methodologies have a goal of identifying data quality issues. An issue is a problem that has surfaced, that is clearly defined, and that either is costing the corporation something valuable (such as money, time, or customers) or has the potential of costing the corporation something valuable. Issues are actionable items: they result in activities that change the data quality of one or more databases. Once identified, issues are managed through an issues management process to determine value, remedies, resolution, and monitoring of results. The process of issue management is discussed more fully in the next chapter.