Data warehousing technology has now been adopted by many organisations as they grapple with the problem of extracting some value from the vast quantities of data locked away in their computer systems. But has technology delivered the improvement in business performance that it promised?
Jon Crews of PA Consulting Group describes an approach to projects which maximises the chances of delivering the benefits.
Surveys show that at the end of 1997, 20 percent of organisations had a data warehouse. Analysts predict that this figure will rise to 80 percent by the end of this year, and the market is estimated to be worth $30 billion by the year 2000. Many data warehouses have been built, many lessons learned, and there are now many tools and theories.
However, the surveys also show that despite the large amounts being spent on data warehousing – with implementation costs averaging £1.25 million – most projects fall short of expectations. Reports show that, even for ‘successful’ implementations, organisations have been underwhelmed as benefits fall well short of those originally envisaged. Where benefits have been identified, they are largely anecdotal. Actual quantifiable results (an x percent increase in converted sales opportunities or n percent reduction in customer churn) are few and far between.
Why do benefits fall short of expectations?
It is the view of PA Consulting that the answer lies not in the technology itself, but in the way the projects have been carried out. Specifically, the shortfall in the delivery of benefits is often a direct result of projects being led by the capabilities of new technology and inadequately focused on the business value.
These problems can be addressed by adopting a more business-driven approach to the development project. This means:
- Defining long-term business goals and developing benefit-based targets for the project
- Actively encouraging business involvement and having a plan to manage the non-technical issues
- Adopting a pragmatic approach for the development to ensure the right technical solution is delivered within short time scales
- Addressing the major data quality issues during the development period.
Defining goals and developing targets
Organisations need long-term quantifiable goals if they are to maximise the benefits from their systems and demonstrate their success. For example, strategic targets defined for a retail bank may include a ten percent improvement in cross-selling capability in the general insurance market. In the initial stages of the project these targets form the basis of the business case and help to determine the funds to be invested.
Having planned the long-term goals, it is important to put short- to medium-term benefit-based targets and business performance indicators in place, against which the value of the data warehouse and the project can be measured. These goals form a basis for maintaining the focus of the project. In the retail bank example these interim project targets may include provision of clean product-to-customer data available though standard reporting tools within three months.
Starting from this point, it is often possible to deliver significant benefit with an inherently simple technical solution. Business benefit-based targets also give a project the impetus it needs to change the organisation or business processes for the delivery of benefits.
Encouraging business involvement and a plan for non-technical issues
For a data warehousing project to succeed, the project team must bring together different parts of the business to develop a consensus and gain buy-in. Failure to achieve either of these will eventually result in political and non-technical issues killing it.
The prime objective of any data warehousing project is to develop a single model of the business and/or the external market. Consensus must be achieved between different groups in the organisation to develop this model:
- The model must be consistent and acceptable to each individual business area
- Different areas of an organisation will frequently make different assumptions about how its business is run
- Exceptional requirements (for example, those specific to individual parts of the business) must be identified and catered for.
Each part of the business involved with the project must take responsibility for its individual data streams. Once the central model is developed, the individual business areas are best positioned to map their ‘raw’ data elements on to the central model. This is because the users are the only people who know how to resolve anomalies in the data and are therefore best placed to decide how ‘dirty data’ elements should be processed during the data clean-up.
Business users must be bought in to ensure that the organisational and process changes exist to support and make use of the data warehouse. The business processes must exist to act on the information the data warehouse provides. Business users are useful allies in sponsoring source data and are often the only ones who are in a position to ensure long-term data quality improvement in source systems.
It is easy for a project to diminish the value of its technical solution by underestimating the amount of effort required from the business community to gain support. A delegate attending a PA Consulting data warehousing workshop at the 1997 IT Directors Forum remarked: "For our project, 20 percent of the effort was technical. The rest was managing the politics and getting the buy-in".
Adopting a pragmatic approach
With data warehousing projects there are many tools that could be used to deliver an adequate solution. However, it is easy to spend a lot of time searching for the ‘best’ tool for the job. For example, a company in the health sector spent £250,000 on licences for an OLAP reporting tool. The development team later discovered that reporting needs could be met with a simple Visual Basic system for a fraction of the cost.
A pragmatic approach is needed to ensure that the final technical platform can support the constraints and demands of the environment.
- Data loading can be a problem if data volumes are large
- Data cleaning will depend entirely on the type of project and the nature of the source systems
- Many tools assume a high-performance infrastructure and fail to perform in more hostile environments.
There are many reporting tools supporting functions such as drill-down, multi-dimensional reporting and traffic light alerting. A pragmatic approach should be adopted to ensure that the reporting mechanism really addresses the required sophistication of reporting analysis as well as the sophistication of the end-users.
Address the major data quality issues
To maximise the benefits from a data warehousing project, the major data quality issues should be resolved during the implementation period. ‘Dirty data’ is the prime inhibitor to the realisation of benefits. The project-based development team are best placed to resolve major issues quickly and efficiently. Sustainable data clean-up procedures are critical for the ongoing survival and value of a data warehouse.
If the data in a system is bad quality, people will not use it. Experience has shown that ‘dirty data’ makes analysis cumbersome for end-users, system results less credible and the system generally unattractive and difficult to use. For example, duplicate data items, such as customer names, immediately switch off users. It only takes two or three duplicates in a list to give the impression that the data in the system is unreliable.
Business users are generally far more interested in the data content of a system than the functionality the system offers. This is illustrated by a group of bankers who were asked to comment on the functionality of a new OLAP system in the early stages of development. After a week of experimentation, the staff felt they could not appraise the system, as the data had not fully reflected their business.
End-users understand their own data, have their own view of the business and expect the system to reflect that view. For example, a group of buyers were asked to test a data warehouse. The system failed because the reports generated by the system did not match the users’ perceptions of their business, who were saying, for example, "I know we do £2 million of business with X, but I can’t see this in your report".
By addressing data quality issues whilst the development is still in progress, a project can often achieve good improvements by applying skills within the project:
- Technical team members can use data base query tools and spreadsheets to update data in bulk on behalf of business users
- Development team members have access to the full picture and are in a position to address cross-business problems such as duplicate customer names
- Development team members are a dedicated resource and able to focus on the job in hand.
Sustainable data clean-up and management processes are critical for the ongoing success and survival of a data warehouse. If the implementation of these processes is left to the end of the project, they will have insufficient business buy-in and momentum, and will not be adopted, leading to the quality of the warehouse data gradually degrading. It is critical then, that these clean-up procedures are implemented whilst the development project is still there to support them and tune them where necessary. When the implementation project finally ends, there must be a critical mass of business users committed maintaining the quality of the system’s data.
It is only by adopting these key principles that an organisation can be confident that the data warehouse under development will ultimately achieve the benefits identified at the project’s conception. It is a sad fact that too many data warehousing projects fail to engage the business in the development process. They lack clear business targets against which to monitor success or failure, get too excited by the technology rather that the required benefits and fail to address data quality issues early enough in the project. This in turn implies that many organisations are failing to fully achieve targeted improvements in key areas such as customer service and retention, sales effectiveness and business process improvement.
