A recent Data Strategy article about marketing databases made a number of interesting points (“360 degree view without going blind”). But ultimately, we would argue, it draws the wrong conclusions.
Detailed data – and lots of it, captured across all channels – is vital to a complete understanding of customer behaviour. For example, one Teradata customer built a customer profitability score from the detailed sales revenue and summarised cost data in its data warehouse. While this application was very successful, when the same organisation later revised this model to include detailed cost accounting data, it found that the profitability scores for 75 per cent of its customers changed by more than two deciles.
Another organisation captured detailed data from its website with the objective of identifying abandoned baskets and incentivising customers to complete these transactions. When they brought this data together with data from their other sales channels, it found that many of these transactions weren’t abandoned at all – they were simply completed via the call centre.
Integrating the data in the data warehouse instantly made it clear that providing further discounts for these high-cost-to-serve customers was absolutely the wrong thing to do. But when the organisation had viewed the web channel data in isolation, that was precisely what it had planned to do.
All of which is not to say that organisations should spend years collecting every available data element and then dumping them all on the desk of every marketing analyst. The most successful data warehouses are delivered subject area by subject area, function by function, in 60 to 90 day increments and with the active participation of both business and IT.
Strong business leadership is essential, but in this the marketing database project is no different from any other critical business project. And any decent analytical database solution should always include a layered data model with a simplified “semantic layer”, so that less demanding users are enabled to access just the data that they need, when they need it.
Successful data warehouses also increasingly include “laboratory” or “sandbox” areas into which end-users can upload their own demographic data, for example, to join them and compare them with the other data already in the data warehouse. In this way, the marketing department can establish within hours whether a new data source is interesting or not and choose only to invest the time and effort required to deliver repeatable, certified loading processes for the data that adds value.
As one of the contributors to the article points out, “you build a single marketing view for political reasons, not because of any technical challenges”. How much better to tackle the politics head-on and incrementally build out a robust, shared and re-useable infrastructure than to deliver a plethora of expensive point solutions.