Plan and Select a Subject Area
When a company first makes an investment in business intelligence technologies, they should plan and prioritize what areas of the business to provide reporting and analytical solutions and what sequence makes sense, both from a business standpoint as well as a technical standpoint. A good example of logical sequencing involves customer profitability. Most companies want to know their profitability by customer, but this typically involves an allocation process that is driven by sales activities and general ledger expenses. Thus, customer profitability can not be delivered until the business subject areas of sales and general ledger have been implemented into the business intelligence solution.

Once a high level plan of attack has been devised, you can then begin implementing each subject area of the business one at a time. This approach provides deliverables to the user community in incremental segments, and it also speeds up subsequent implementation cycles, as previously extracted components will get reused, such as the customer dimension.

Gather Requirements
Even though gathering requirements sounds straightforward, this is the area that companies tend to oversimplify and are often not thorough enough on the initial pass. This can cause redesign and modifications prior to going live, which can jeopardize user confidence in the project and negatively impact the timeline.

The first step is to interview a cross section of end users (both upper and lower level employees) for their reporting needs. Once these requirements are understood and documented, the next step is to interview appropriate technical resources to understand where the source data will come from, and if there are any known data issues. It is quite common that data issues, such as inconsistent use of a field, will bring the team back to the business users to see how they want to handle certain exceptions. Once all data issues are identified and resolved, the business requirements should then be signed off by the user representatives and the design phase can be completed.

Design
There are several areas of design to be completed in a typical business intelligence project. This includes the design of the relational data warehouse, the design of the ETL process required to load the data warehouse, and the design of any specific OLAP data marts, such as Essbase or Analysis Services "cubes." Although this step is listed after the requirements phase, it is common that design drafts will begin prior to the completion of the requirements phase in order to expedite the implementation process.

We are advocates of a star-schema approach to the enterprise data warehouse design. This provides ease of use and performance gains for end user access and data retrieval. Ideally the data warehouse designer will have skills in both the ETL tools being used, and the OLAP data mart technology being used. Having these complementary skills in these different areas accelerates the final design, prevents rework, and optimizes the load process performance. Analytic Vision believes strongly in cross-training their employees to optimize this important phase of the project. Read more about our recommended Dimensional Modeling approach in the Services section of the web site.

If a company is implementing a quick prototype with a data mart tool, it is common for the data warehouse step to be initially skipped, but we strongly recommend that companies implement an enterprise data warehouse as they move forward with their business intelligence solution, to guarantee consistency of terminology and data values across all data marts. A properly designed data warehouse will improve performance and subsequent implementations, as your business intelligence solution grows.

The design phase also includes the design of any data marts and a base set of expected OLAP views and reports. The design and view mockups are part of a review session that allows the user one more shot at providing feedback toward the expected solution. This step not only sets the user expectations of the deliverables, but it is also important to make sure that the users feel like they are part of the solution, which is extremely helpful during the deployment phase.

Develop
Whether your team is using a high-end ETL tool such as Informatica or DataStage, or development tools such as Microsoft’s DTS and TSQL, there are many common data transformation routines (such as using cross-reference tables and performing allocations) that will have to be utilized. At Analytic Vision we leverage our experience to understand what techniques offer optimal performance, and we are very adept at developing custom code routines for dealing with complex data issues. Read more about our recommended Back End Automation approach in the Services section of the web site.

Besides developing the ETL processes to load the data warehouse, it is extremely important to test the processes for accuracy and speed. We believe that the developer should understand what the expectations are for clean data, and that they ideally can identify unexpected data issues that sometimes are discovered when dealing with the real data. To assist this process, team members from the user community should be involved in final data validation. We recommend developing audit queries that automatically tie the data warehouse data back to the source system whenever possible.

User Review and Training
Prior to deployment of the solution, it is important to provide a detailed review session to get any final user feedback and ideally to generate user enthusiasm for the solution rollout. After the review is done and any final modifications are complete, we recommend that some organized training with handouts be provided to the user community. Many projects fail because the users are not confident in the use of the deliverables (which may seem really easy to IT personnel). We recommend developing a static training data mart that has a training guide with "how to" exercises that users can take back to their desk after the training class. This provides them with examples of their typical tasks that they can reference anytime they need.

Deploy
The deployment phase includes tasks such as setting up appropriate security, implementing backup and recovery procedures, and automating the routine data warehouse refresh processes. Some key users should be identified for scrutinizing the initial data loads to make sure no issues arise. Having user representatives vouch for the data will provide confidence to the general user population as to the quality of the information.
Automated audit processes should be utilized to check some key figures between source system, staging area, data warehouse, and data marts when possible. Performance and data quality should be monitored after deployment, and help line procedures should be put in place for assisting users and tracking feedback for enhancements or changes.