A colleague of mine has recently tried to convince me that (a) Lifetime Value (LTV) is a meaningless statistic and (b) applying prediction science to marketing will result in the same failures as those experienced by the stats-heavy financial industry.
While I agree on the first point (whose lifetime we talking about, anyway), I think the second is an over-reach.
You certainly can't predict everything - especially the future with any degree of certainty. But saying that you can't predict ANYTHING seems irrational. There's a business that lives and breathes this uncertainty every day - the insurance business.
Insurance companies have no idea when a catastrophe will happen. But you can use the principle of the law of large numbers to mitigate the risk associated with an any unforeseen event. The key learning from the law of large numbers is that you can't always predict the outcome of the individual (too much variability) but - if you have enough of exposures that look alike - you can better predict the results of the group. Not always perfectly, but better. And there's value in better. Often, significant value.
It's not perfect, but it works. You can think whatever you want of your insurance company and point to the AIGs with their over-reliance on stats and mortgage securities got them into into serious trouble. But you also have others that have been doing this sort of "prediction" for more than 100 years.
Here's a really dumb example. If I punch any one of you in the face, I have no idea what the outcome will be. In fact, if I punch one of you and nothing happens, I may conclude that there are no repercussions to the act. But if I punch ALL of you in the face (and live to tell about it - my readers scare me) I can come up with a range of possible outcomes and "predict" an overall impact to my well being if I find another group of similar people and punch THEM all in the face.
Here's another, more relevant, example - Hurricane Irene. Models and prediction tend to break down the more finite the number of observations. So while you can say things like "there are 1.5 hurricanes per year on average", you can't predict when Irene will come. Or where it will go when it will gets here. But you can "predict" that Atlantic coastal states have a significantly higher chance of a hurricane than Colorado. You can also "predict" what happens when six feet of water come rushing down your street. You can also "predict" that houses closer to low-lying areas are more likely to flood than ones high on a hill. So why you didn't know that Irene was coming, you could defend against the event better than if you hadn't examined and learned from the past. And there's value in better.
In business (and especially email marketing) we are more concerned with the behaviors of the group more than of any one individual. So why do we do this? Simple...scale. We're not mom and pop marketers. We have a lot of large numbers from where we can draw conclusions. When we try to capture data about a person, it's so we can better associate them to a group (a cluster - hooray for clustering!) whose behavior - driven by the law of large numbers ' we can better predict.
It's certainly not perfect. But it's not value-less. There's a huge aspect of time-frame management that needs to be considered in terms of setting accurate probabilities. You can learn a lot from watching and mining the behaviors of your customers. You can apply that learning to prospects. You can use the law of large numbers to drive better, more efficient value from your marketing efforts. Really, that's all email marketers do. We're not here to shape the future direction of the business - we're here to make money while there's money to be made.
So while I agree that LTV is a bit like finding Noah's Ark, to say that prediction is a losing proposition swings the pendulum too far to the other side. It's not the prediction that's the problem - it's the incorrect use of the tool.