IBM Planning Analytics – Characteristics of a “Good” data model

Modeling

Many “strategies” exist today for modelling (building a database) in IBM Planning Analytics. If you are a member of the IBM Community, you’ll find no shortage of opinions and recommendations – from informal blogs and discussions to more formalized courses and mentoring.

In fact, there is a free and downloadable “educational guide” that covers an overview of the steps involved in building a model including the details involved in each step.  The guide is available (in the IBM Community library) here: https://community.ibm.com/community/user/businessanalytics/viewdocument/building-a-model-in-planning-analyt (note: to supplement understanding of the material, you can also download the database which is referenced in the guide and use it to explore the modelling objects in more detail).

Beyond the Build

The Planning Analytics Workspace (PAW) platform can really simplify creating a data model – still, as an architect, you should take time to carefully design your data model, making sure that it aligns with current requirements as well as the application’s expected future state.

A “good” design makes it easier to scale your model, maintain it, as well as take advantage of current and perhaps future Planning Analytics features, and of course better serves the needs of the business, while the impact of poor design decisions made when designing a model are often not immediately noticeable. Rather, the “weakness” of a model is exposed as the application scales.

“Scaling” can include adding additional users, modifications to security schemas, increasing data volume or velocity, material data reclassifications, and so on.

The Perfect Design

The first step to beginning a data modeling exercise is to accept that there are no “perfect” data model designs, only very good ones. Since organizational (and data) requirements grow and evolve, there can be no “one size fits all” or “once and done” models.

 

A “good” data model design is one that serves as a foundation for data use by providing correct answers from trusted data sources, can “flex” to handle changing data formats and needs , and is user-friendly in that it is easy for users to view and understand the data and its relationships.

 

Some other characteristics of a “good” data model design include:

  • Limited errors in development
  • Data can be easily consumed by all
  • Acceptable and Predictable levels of performance
  • Flexible to meet changes in requirements
  • Good naming standards
  • Standardized data practices
  • Reasonable levels of data duplication
  • Easy and robust validation
  • Is extensible and reusable

Next Steps

With the above “top of mind”, how do you ensure that the model design that you came up with will be “good”?

Firstly, you must thoroughly understand the objective of the new model (why are you building a new model?) and then set specific goals for the design (the more specific, the better) to meet those objectives or to “satisfy the “why”. Finally, it is always helpful to have a seasoned expert perform an architectural review of your proposed design or, if you have a model already up and running, conduct a health assessment. If you are interested in learning more, QueBIT Advisory Services provide guidance for organizations in many ways including these activities.