If you're attending DCI's show in Chicago next month, be sure to stop by. Details for my session are on the right side of this newsletter.

And, if you're attending Gartner's Business Intelligence Summit March 6-8 (also in Chicago) I'll be in the Solution Showcases Monday and Tuesday, and delivering a session on Monday.

Be Prepared to Duel with Data Quality

by Rick Sherman, Athena IT Solutions

Plenty of business intelligence or data warehouse projects have been blindsided by complications related to data quality. Sometimes these issues aren't apparent until business users start testing the system just before going live with the project. So what causes BI project teams to get caught off guard by data quality issues? And why do these problems surface so late in the project?

There are two common pitfalls: defining data quality too narrowly and assuming data quality is the responsibility of the source systems.

People often assume that data quality simply means eliminating bad data -- data that is missing, inaccurate or incorrect. Bad data is certainly a problem, but it isn't the only problem. Good data quality programs also ensure that data is comprehensive, consistent, relevant and timely.

Don't blame the source systems

Defining data quality too narrowly often leads people to assume that source transactional systems -- either through data entry or systemic errors -- cause the bad data. Although they may be a source of some errors, the more likely culprits are

either inconsistent dimensions across source systems (such as customer or product identifiers) or inconsistent definitions for derived data across organizations. Conforming dimensions -- developing consistent customer or product identifiers -- is important for accessing and analyzing data for a company. The source systems do not own the data quality issues across other systems, the BI project team does. The source systems need to ensure that the data within their data silo is correct. But the BI project team is responsible for providing the business with data that is consistent across the enterprise.

Similarly, each organization within the enterprise may have valid business reasons to derive data differently than others. For example, their position in a set of business processes may determine how they view their data. The individual organizations aren't tasked with developing common definitions for derived data, but the BI project team is. Many BI project teams try to claim that data quality issues aren't their responsibility. However, from a practical viewpoint, the BI team does need to make these issues their own, since their job is to ensure the highest data quality possible. The BI project team is packaging the data for consumption by business users and they will be held accountable for the data quality. This may not seem fair, but the success of their project depends on it.

Don't shortchange the pilot

Surprises happen when the project does an initial pilot or release involving only a small subset of source systems. While there may be many good reasons to have a narrow scope for a pilot, you won't get an appreciation for the effort necessary to conform these dimensions as the number of source systems expands.

Sometimes pilots are only with a single organization, using only their definitions for derived data. Once again, the tough issue is often how to accommodate the differences in the derivation definitions between organizations. In both cases the real challenges are encountered when dealing with multiple systems and organizations. The business users need to look at the big picture, and that is only possible when they can access and analyze data across the enterprise.

Steps to address data quality

To ensure data quality, the BI project team has to address it from the very beginning. Here are several significant steps to consider:

1. Require the business to define data quality in a broad sense, establish metrics to monitor and measure it, and determine what should be done if the data fails to meet these metrics.
   
   
2. Undertake a comprehensive data profiling effort when performing a source systems analysis. Data anomalies across source systems and time (historical data does not always age well!) is needed so that the team can address them with the business early on.
   
   
3. Incorporate data quality into all data integration and business intelligence processes from data sourcing to information consumption by the business user. Data quality issues need to be detected as early in the processes as possible and dealt with as defined in the business requirements.

Enterprises must present data that meets very stringent data quality levels, especially in light of recent compliance regulations and demands. The level of data transparency needed can only result from establishing a strong commitment to data quality and building the processes to ensure it.

For more information on training, see our page on data warehouse training.