• This field is for validation purposes and should be left unchanged.
Guidance and results for the data-rich, yet information-poor.

All Models are Wrong, but Some are Useful

CloseEnoughThe famous statistician George E.P. Box once said, “All models are wrong, but some are useful.” That is a wonderful statement to remember as your organization approaches predictive modeling and data analytics.

You see, machine learning is excellent at a lot of things, but you would never want to use it to balance your checkbook. That’s because predictive modeling usually gets “close enough” to be useful. Ballpark predictions can be incredibly helpful.

For example, let’s say you’re trying to figure out how to sell more raincoats. Your model might tell you that you sell more raincoats when you place them next to the umbrellas. You’re not going to get an accurate figure–the model will not tell you how many more raincoats you can expect to sell. The model can, however, tell you that you will sell more raincoats, which means you’re going to see a benefit from that model that you would not have seen if you had not taken the time to find that correlation.

In order to understand why it’s not that important to get a “perfect” model, you need to understand the concept of incremental enhancement.

Here’s another example. You’re doing a direct mail campaign. You get a 2% return, which means you’re wrong 98% of the time.

You run a predictive model which tells you how to get a 3% return on your direct mail. That 3% return is actually enormous in terms of dollars spent, and you’ll receive a significant ROI from that little, incremental enhancement. And you’re now only 97% wrong.  You can, perhaps, eventually tighten things up even more so that you’re only 96% of the way wrong. And each of these little increments represents a phenomenal amount of return with tremendous value for your organization.

As it is, there is no model that can perfectly predict human behavior. Human behavior is messy. It’s inconsistent. But you can certainly get “close enough” to make money.

If you’d like to find out how you can build useful models that give your organization to high-value incremental advancements, then it’s time to register for TMA’s free webinar: Data Mining, Failure to Launch. You’ll get a solid understanding of some of the biggest mistakes that organizations make when they try to start using data mining and predictive analytics for the first time, so that you can avoid these pitfalls and see real returns on your data mining investment. Register now!



Data Mining Webinar

Learn How to Get Predictive Modeling
Off the Ground and Into Orbit
1 Hour Live Interactive Event

Why Train With TMA?

Determine whether TMA training is right for you, and learn why TMA is truly the best option for live classroom analytics training.