By way of Multichannel Merchant, we have this article: Call it E-RFM
by Ken Magill. His idea will probably be a stretch for many in the e-mail space, but it’s a great example of what I was talking about in How to do Customer Marketing testing.
The gist of the article is you can reduce spam complaints and better manage reputation by anticipating which segments of subscribers are going to click the spam button. Yes, anticipate. You know, predict?
For some reason, online marketers seem like they are not really into the prediction thing – or at least are unwilling to fess up to it. Test, measure, test, measure, web analytics is mostly about history, as opposed to predicting the future. How about predict, measure, predict, measure? Same thing, only much more powerful – if you can guess what customers will do before they do it, you have real marketing power. Perhaps this is why folks don’t talk about it much…
The prediction model discussed, RFM, is one of the most durable and flexible models in the entire BI quiver. As Arthur Middleton Hughes says in this article, “There isn’t a predictive model in the world that doesn’t have RFM inside of it”.
And the RFM model is free. You don’t even have to hire a statistician!
The RFM model sometimes gets a bad rap because people use it with very little imagination, simply reproducing the basic catalog model from the 1950’s, instead of understanding the guts of it and using it in new ways. This Call it E-RFM article is a good example of how to use RFM in a new way; a broader explanation of using modified RFM for e-mail is here.
Those of you interested in how to really take advantage of the new Webtrends Score product should pay attention to this prediction area, because “Potential Value” – a prediction – is absolutely fundamental to optimizing a Score model. You could use Score to predict which segments are most likely to click the spam button. And then you could test, track, and fine-tune those predictions until you get them right. Sounds like fun, huh? Does to me, anyway…
But you don’t need something like Score to predict likelihood to click the spam button; sending an e-mail every week for 3 years to somebody who never clicks through should be a rough indication…
So, do you use predictive models in your work? Why or why not?
If you don’t use prediction, is it because coming up with a great campaign for a prediction is the problem? Or because nobody really cares about customer marketing, it’s all about customer acquisition?