Efficient Databases Require Past-Point-in-Time Views

Jul 07, 2008 11:16 PM  By

Most marketing database developers are not deep-dive data miners. Therefore, all too many of today’s databases don’t support strategic analytics best practices.

An important factor is a misunderstanding of time as it relates to data mining. Analytical professionals work in the present, on the past, in anticipation of the future. Therefore, a seminal requirement of a marketing database is the easy and rapid recreation of past-point-in-time views.

Past-point-in-time views are essential for the construction of the analysis and validation files required for predictive models. Likewise, they support the creation of the underlying data required for all cohort analysis, including long-term value and the monitoring of changes in customer inventories such as fluctuations in segment sizes over time.

Past-point-in-time views also are known as “states” or “time-0 views.” A third common term, “freeze files,” refers to archaic database processes. The following test will help determine if your firm is using the best practices of marketing databases:

  • Consider a promotional sequence that is done within a marketing database environment—say, a mailing with a follow-up e-mail to your best customers.
  • Assume you intend to perform subsequent deep-dive data mining on this promotional sequence—perhaps, a predictive model or a sophisticated response analysis.
  • If, in conjunction with the database selection (“pull”) to execute the promotional sequence, you have to create a “freeze file” so you can do the analysis later, then you are not working with a best practices marketing database. This is because all the data required for deep-dive data mining – by definition – already resides within the database.

Jim Wheaton is a co-founder of Daystar Wheaton Group.