Tag Archives: prediction models

Database Marketing Metrics in a Digital World

I have had a lot of conversations in the last few months about the fact that digital marketing is fundamentally not that different from traditional marketing.

Of course important adjustments are in order when you have to communicate to a consumer who is in control and has less tolerance than ever before for interruption and irrelevant messages. But a lot of things you know about marketing stay the same, and digital only brings more depth and richness to the way. Here is one example from the world of database marketing.

In the early days of database marketing, catalogue marketers were trying to find ways to identify which customers were most likely to buy than others when there was a communication effort. The straightforward way was to segment customers based on how much they had spent to date – and that turned out logically to be a good predictor (or “proxy” for the DM purists).

But relying on the monetary value only proved somewhat limiting: there were lots of customers who were buying even if they were amongst the most valuable ones, and there were many customers identified as very valuable who used to buy a lot but not anymore (you may have heard of them as “lapsed customers”).

It turned out that by looking also at how recently and how frequently customer bought over the last few months, the accuracy was greatly improved. By adding and combining these two additional parameters, database marketers could guess who would buy and who would not – and that could lead to significant savings in printing and postage costs.

And so very empirically the RFM prediction model was born – R for Recency, F for Frequency and M for Monetary value or Money spent. Its efficacy is still as impressive today as it was over the past two decades. Of course, more sophisticated models have appeared over time using statistical techniques such as regression analysis, neural networks or genetic algorithms… RFM has the merit of simplicity and can be applied without advanced statistical know how and resources.

Visit this site if you are interested in reading more about RFM.

So how do you apply a digital lens to RFM? The model stays the same but there is room for additional information. RFM appeared in an era where storage and the number of opportunities for collecting data were limiting factors. RFM focuses therefore on transaction data – but does not necessarily leverage interaction data which can be more easily tracked today.

There is an opportunity to look at two more dimensions: Attention and Engagement. Attention can be for example measured by how frequently a consumer opens your emails or how many of your last 10 emails they opened. Engagement can be indicated by the time a consumer spends on your web site after clicking from an e-mail. In fact the model can be extended outside of e-mail marketing to include online advertising – although it would apply in this case to anonymous prospects rather than identified customers. And in a not so distant future, with new technology that allows delivering targeted TV advertising to specific households, such an approach could expand to other marketing channels.

Moving from RFM to FRAME is a good way to describe how marketers should approach their migration to the digital world – key basic principles stay the same, and additional data open up new opportunities for understanding customers. At the same time, it is very easy to look at too much information or at the wrong information: so when it comes to new marketing models, keeping things simple is paramount.