Predictive Analytics Gets Closer

I am always a little shocked by a company’s resistance to using predictive analytics. My guess is that is a combination of not really understanding the value, fearful that they won’t get it right, or not having the right talent to use it. It has long been labeled as “white lab coat stuff” and perhaps that is a bit accurate. But software is making this easier, and MBAs are studying it so this label should be diminishing.

The Value: Reducing costs, increasing returns, quicker identification of issues – these are all critical wants of every organization. If we can only chase five opportunities with roughly the same make up, a little predictive analytics should be able to tell you who is more likely to have a higher customer lifecycle value. If you only can cover 10% of the market with a marketing campaign, predictive analytics can help you determine which 10% is likely to have the greatest yield.

The Fear: I understand this, but it is a little irrational as all decisions involve some level of risk. All predictive analytics do is make decisions based on an elevated likelihood of being right. If I told you I could make you 10% smarter, wouldn’t you listen?

The Talent: This is …

The Talent: This is perhaps a realistic barrier, but one simply corrected. Predictive Analytics, while getting easier every day, is still about advanced computations. Not only do you need to understand how to do them, you need to understand when and to use them. And more importantly, you need to understand how to transform the information into values an executive team can put into action.

Where do you begin:

Find someone in the organization with a good statistical and business mind (or hire one). This may not be the technical team – it often takes a little different skill set. Or find a small team.
Find a business process where there is pretty good data and that will add value at the end of the day – customer attraction, attrition, fraud detection, scrap reduction, etc.
Put a small project in place to try it.
Enter my favorite stats words – Parsimony: Find the most simple answer. This is easier to explain and digest of how to put the project into action. (Why is a word that strange about the simplest answer). It is easy to end up tweaking a project to death. Don’t do it on the first pass. You get lost in data and often find it far more difficult to explain and complete the project.
Try it and accept the results. The is tremendous learning in failing (and chances are likely you won’t fail if you didn’t bite off that much).

Examples:

Let’s say you can identify customers who are likely to abandon you and then work to make sure those customers are treated better. If your abandonment rate drops by 10%, what is the value to the bottom line?
If you can identify customer segments that are less price sensitive, what is the value of a 1% increase in average deal size (note that the entire amount really should drop to the bottom line as well)?
What if you can reduce fraud by 5%?

The numbers show that predictive analytics are very real. It is not about guessing, it is about reducing the risk of guessing. And if you follow many blogs, all of a sudden there is a lot more information on predictive analytics. IBM is finally putting together some wood behind the arrow of its SPSS purchase which may also begin to influence more decision makers in the space.