Mega-Analytics: Factor Analysis

If you're a multichannel analytics guru, it is very likely that you've learned an awful lot about customer behavior by running a procedure called a "Factor Analysis".

At a simplistic level, a factor analysis takes a bunch of variables that are highly correlated, and transforms them into variables that are not correlated.

For instance, your "best" customers all have common attributes.

1. They all purchased recently.
2. They purchase all the time.
3. They tend to buy from multiple channels.
4. They tend to buy from multiple merchandise divisions.
5. They tend to buy across seasons.
6. They tend to buy full price and sale priced merchandise.
7. They tend to pay for shipping and take advantage of free shipping.

All of this "multi" behavior becomes hard to understand. The factor analysis makes this a lot easier.

When I run a factor analysis for my clients, I tend to observe the following trends.

1. All multi behavior ends up being combined in the first factor.
2. Unique channel activities end up being separated out in subsequent factors. You'll see a factor that focuses on online customers who only buy when a free shipping promotion exists as a unique factor, for instance.
3. Unique merchandise activity ends up being separated out in a subsequent factor. You'll see customers who purchase iPods, and they are fundamentally different from the customer who purchases a Blu-Ray DVD player.

Once you have identified each factor, you run your typical logistic regression (response) and ordinary least squares regression (spend) models. Now, you truly know what attributes are driving your business! The factors that are statistically significant are the ones that matter.

I like to take the top 2-3 factors, and segment customers into eight or nine combinations based on the values of each factor. Then, I watch how customers migrate between the segments, over time. You'll notice trends like catalog ---> e-mail ---> e-commerce ---> retail happen when you evaluate migration across factor segments.

So sit down with your analytics expert, and run a factor analysis with all of those highly correlated variables in your database, looking to identify the factors that truly drive your business.