Success factors of implementation

Amount and quality of available data

Without proper data, or with too little quality data, any BI implementation fails; it does not matter how good the management sponsorship or business-driven motivation is. Before implementation it is a good idea to do data profiling. This analysis identifies the "content, consistency and structure [..]" of the data. This should be done as early as possible in the process and if the analysis shows that data is lacking, put the project on hold temporarily while the IT department figures out how to properly collect data.

When planning for business data and business intelligence requirements, it is always advisable to consider specific scenarios that apply to a particular organization, and then select the business intelligence features best suited for the scenario.

Often, scenarios revolve around distinct business processes, each built on one or more data sources. These sources are used by features that present that data as information to knowledge workers, who subsequently act on that information. The business needs of the organization for each business process adopted correspond to the essential steps of business intelligence. These essential steps of business intelligence include but are not limited to:

  1. Go through business data sources in order to collect needed data
  2. Convert business data to information and present appropriately
  3. Query and analyze data
  4. Act on the collected data

The quality aspect in business intelligence should cover all the process from the source data to the final reporting. At each step, the quality gates are different:

  1. Source Data:
    • Data Standardization: make data comparable (same unit, same pattern…)
    • Master Data Management: unique referential
  2. Operational Data Store (ODS):
    • Data Cleansing: detect & correct inaccurate data
    • Data Profiling: check inappropriate value, null/empty
  3. Data warehouse:
    • Completeness: check that all expected data are loaded
    • Referential integrity: unique and existing referential over all sources
    • Consistency between sources: check consolidated data vs sources
  4. Reporting:
    • Uniqueness of indicators: only one share dictionary of indicators
    • Formula accuracy: local reporting formula should be avoided or checked