How Marketing Databases Differ from Operational Databases
A marketing database must be able to perform all of the mission-critical analytical tasks required for data-driven marketing. Many people think they have a marketing database when, in reality, what they have is an operational database. An operational database supports essential "nuts and bolts" tasks such as customer service, fulfillment and inventory management. But, it falls short in the support of data-driven marketing analysis.
To determine if you have a marketing database, take the following data processing test. If you can easily and rapidly execute the five tasks within the test, with no outside-the-system processing, then you will know for sure that you have a marketing database:
1. Examine the life-to-date detail for your customers as of June 1, 2009; that is, about a year ago. This is known as a past-point-in-time ("time-0") view, which will be impossible to recreate if any of the following is true:
· Some of your customers as of June 1, 2009 are no longer in the system.
· Some of the historical data previous to June 1, 2009, for some or all of your customers, has been deleted or overwritten.
· You cannot exclude from your examination all historical data subsequent to June 1, 2009.
2. Rank your customers from best to worst, as they would have been ranked on June 1, 2009. Do this by evaluating each customer's year-ago view by whatever selection system you use; that is, a statistics-based predictive model (or models), or some sort of rules-based logic such as Recency/Frequency/Monetary ("RFM") cells.
3. Divide the ranked customers into deciles; that is, into equal groups of 10.
4. For each decile, calculate the following subsequent performance; that is, from June 1, 2009 through May 31, 2010: Average Per-Customer Revenue and Average Per-Customer Promotional Spend. Please note that the second will be impossible to calculate if you do not maintain all-important promotion history for all your customers on a campaign-by-campaign basis, regardless of whether a given customer did or did not respond to a given campaign.
If you can do all this, then you might have a marketing database. To know for sure, you need to be able to do one last thing:
5. Simultaneously for each of three additional past-points-in-time – that is, June 1 for each of the years 2008, 2007 and 2006 – create a standard file inventory report. The specifics will vary by the type of business you are in, but invariably will include:
· Permutations of customer counts, purchase rates and dollar amounts, and
· Year-over-year absolute as well as percent changes.
Components of your file inventory report should also double as key performance indicators ("KPI's") that are closely tracked throughout the organization.
If you can do all this, then you really do have an environment worthy of being called a marketing database. The reasons a marketing database needs to be able to do these five tasks are because:
· Database marketing is, by definition, driven by deep-dive data mining.
· Deep-dive data mining, in turn, requires the ability to rapidly recreate past-point-in-time ("time 0") views, and then manipulate and report on the data within these views. In fact, it is common for multiple such views to have to be simultaneously recreated.
· Without this ability, you will not be able to efficiently execute any cohort analysis such as lifetime value. Nor will you be able to easily construct any statistics-based predictive models.
Whether or not the marketing database and the operational database should be the same physical resource is an entirely different issue. And, an entirely different article.
Jim Wheaton ([email protected]) is a principal at Wheaton Group.