Teradata

So, your company has a data warehouse. It was most likely implemented based on a specific business case addressing a specific business improvement opportunity. Everybody has to start somewhere.

What makes an enterprise data warehouse extremely valuable to your business is when your users start gaining new and actionable insight (information/knowledge/wisdom) not previously available. So why don't we know what the insight would be beforehand? Why is it so difficult to predict and quantify benefits? More importantly, what can your business do to get smarter in this area? Teradata Corporation has explored this question. The result is a method of visual modeling to support understanding and planning as described in this white paper.

Figure 1 illustrates companies that have all evolved their data warehouses and grown them to support other uses. Many of them are enterprise data warehouses, spanning multiple areas of the business. 

Good planning has facilitated evolution of their data warehouse as it traveled the route to endless new and actionable insight for the business. This visual model is called the Teradata® Enterprise Data Warehouse (EDW) Roadmap.

Teradata currently has six EDW Roadmaps developed. They support the retail, communications, banking, insurance, travel, and manufacturing industries.

A two-word title: EDW and roadmap. Many have used the term EDW, but not always with the same meaning. For the purpose of this paper, the general concepts supporting EDW are that it is a central repository for significant parts of an enterprise's data, and that its use extends beyond a specific department or single group of users. Also, its construction tends to be iterative and constantly evolving for new uses.

The term roadmap is borrowed from our love of the automobile and the open highways. With the exception of that small group of males who never need directions, most of us are well aware of the time and money that can be saved by having a resource that can assist in getting us to our destination, while leaving the exact route flexible to change.

'Store once and use many' is a concept Enterprise Data Warehousing strongly supports. The intent is that once data, also called business facts, are stored in a central repository, they can be made available for multiple uses. Two benefits come from that action. One is that the cost of harvesting and cleansing the data is only incurred once. The second is that a single view of the enterprise can be supported. Once a critical mass of data from various sections of the business is combined, new insights (information) about the business also become available. These insights tend to be in addition to the initially planned uses; they are freebies. And they lead to the new and actionable insight that fuels competitive advantage and great value.

A problem with all of this is that building an EDW can be expensive and challenging. The business value analysis becomes complex, as most projects have one business case and, therefore, one associated return on investment (ROI). How can we take the many projects targeted to produce actionable information and then combine and align them with the goals and objectives of the enterprise? And if that could be done, how can we separate that value from the additional value of previously unidentified uses, based on the EDW approach?

While it is complex, expensive, and challenging, it seems reasonable that the rewards justify the challenge. But how can we get smarter about aligning the challenge with the payback? This is the focus of the EDW Roadmap. The balance of this paper describes a model built to illustrate the concepts and values that an EDW can address. The roadmap contains the routes and associated food chain interconnections. The path chosen is based on goals and perceived business values. And, of course, change is inevitable. Therefore the ability to use a roadmap and then model the impact of change will help you maximize return on your EDW.

How is the EDW Roadmap constructed, and how can it build the linkage from the highest strategic levels of a company down to the basic business facts as captured in operational data? The following description is a high-level overview of the roadmap and primarily describes the linkages. The actual model carries much more detail within the model object properties and business rule relationship definitions.

In Figure 2, the upper left most block captures a typical customer’s vision, the linkages to the goals for achieving that vision, and strategic objectives linked to the goals they support. A relative weight is identified for each goal. The weight of the goal is then used to calculate a weight for the strategic objectives. The strategic objectives are decomposed into more
granular Business Improvement Opportunities (BIOs), which identify specific areas of the business where additional business value or business impact can be achieved from the EDW.

The strategic objectives in Figure 3 are linked to the appropriate BIO Objective (12 o’clock position) in Figure 4. The weighted value is carried to the BIO Objective, showing alignment to the company’s strategic objectives. The BIO Objective also captures the business value from the Results component (9 o’clock position) of the BIO. The Results component is based on a
Business Impact Model(s) as appropriate. Knowing the strategic value and business value of a BIO leads to more in-depth understanding of its impact to the business and more informed planning for the EDW.

* Patent pending on the Enterprise Data Warehouse Roadmap Model

A BIO consists of four parts as shown in Figure 5:

• Objective(s)
• Analysis on detailed data to convert it into actionable information
• Business actions based on that information
• Results
  
  
  

A primary assumption of this model is that incomplete sources of data will cause less than perfect production of actionable information. So incomplete analysis leads to less than perfect actions, and, therefore generates less than perfect business value results. As the data sources are enriched, the information improves, supporting higher value/lower cost actions, generating higher value return. This is referred to as 'Information-Enabled Value'.

The model determines Information-Enabled Value through linkages from the Analysis component of the BIO to the data sources. The first link in this path is from the Analysis to Business Questions (BQs) and Key Performance Indicators (KPIs), as shown in Figure 6. They are related to the supporting data identified in the bottom row of the model, in the Logical Data Model (LDM). The assumption is that unless all of the needed attributes are sourced, the BQ or KPI is not valid. For example, a KPI equation with a null numerator or denominator is at best wrong and at worst misleading. Therefore, the BQ or KPI is not considered information of value unless 100% of the needed attributes are available. In the model, a BQ or KPI that is not fully sourced is shown as red. The BQ and KPI objects remain red until all of the needed attributes are noted as sourced. The exception to this is a third analysis type called the business Analytic. An example of this is an analytic algorithm used to do propensity scoring. For this object type, a gradient scale is used to show the value of data enrichment. It is not unusual for a Business Analytic to have many potential attributes in its superset. For example, to address churn, 100 or more attributes can be mined to find the ones pertinent for the building of a churn propensity-scoring algorithm. The value of the analytic increases as the volume of available data attributes increases. In the model, the Business Analytic's associated color changes to reflect the richness of data for the analytic, changing from red to yellow to light green to dark green.

From the business perspective, the LDM is structured as a set of subject areas, represented in the model as the orange containers on the bottom row of Figure 7. A business person usually relates well to these structures. Further decomposition of the subject area exposes the entity and attributes that the IT staff needs to associate with the physical data model and the related operational systems supplying the data. LDMs may contain in excess of 3000 data attributes depending on the maturity of the model.

We have traversed the model from the enterprise goals and objectives to the associated BIOs to the supported Business Questions and Key Performance Indicators and the specific data attributes needed to enable them. Another way to look at the model is from the bottom up. By understanding what data are readily available from the enterprise operational support systems, you can identify what actionable information can be created. The concept of a roadmap is based on incrementally sourcing (loading) operational data into the integrated data model that has been developed for the EDW. The roadmap allows you to make informed decisions on the proper priority and sequence of the candidate source data. As a critical mass of integrated data evolves, exponential return can be expected as multiple departments begin to leverage the business facts, applying the concept of 'store once and use many.'

If you're traveling the EDW route, a good starting point for learning and understanding the way is with an EDW Industry Roadmap from Teradata.