individual-entry-BA-Blog

« Book Review: The Real-Time Enterprise | Main | Insurers should collaborate to fight fraud - with data »

Customer Analysis when you have millions of customers

This interesting article on Customer Flashcards caught my eye yesterday. Chris and Zach do a nice job of presenting the idea of customer flashcards, combining lots of information into a single image, and how that might help you see more complex patterns and recreate a feel of customer intimacy (about which I have blogged before). I really liked the concept and their examples were fascinating. What made me pause, though, was thinking about how to apply this kind of approach when I have hundreds, thousands or even hundreds of thousands of customers. Am I really going to scan that many flashcards? The discussion of Blink was also interesting but again made me wonder about applying this kind of "thin-slicing" (at which humans excel) to huge datasets (something at which humans do not excel).

So how could you apply this kind of approach to huge numbers of customers? It seems to me there are two ways:

  • Firstly you could apply the approach to customer segments. After segmenting your customers using a data mining approach, you could then display the various aspects of each segment using the flashcards. This might help turn the statistical view of the data mining tool/analyst into a more business-focused flashcard, enabling the business folks to better understand the segments that were statistically significant.
  • Secondly you could take the approach of using flashcards as part of identifying the kinds of variables that might make sense in a predictive scorecard. For instance, a cardholder risk scorecard will typically use information like credit history trends and patterns as part of the data analyzed to produce a risk score. This essentially automates for a system the multi-variable/patterns/trends combination presented so nicely in a flashcard.

Like all visualization techniques this has great potential for helping people make decisions. Equally, like all visualization techniques, it becomes challenging when your volume gets high enough to need decision automation so that, for instance, your website can do as good a job deciding as the people who can see the flashcards..

First time on the EDM blog?
Subscribe to the EDM blog feed or check out some other recent posts:

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d83451629b69e200d834cde50569e2

Listed below are links to weblogs that reference Customer Analysis when you have millions of customers:

Comments

Zach

James,
Thanks for the call out and your thoughts on customer flashcards. I'm glad you raise the question of quantifying the visualization. It was something we skimmed over in our article both due to lack of space and because it could get technical (= ?).

True enough, you can't be spending all your time looking at thousands or millions of images (though our main point is that looking at a lot of these is the critical starting point). And if you want to segment your customers by behaviors, you do need to analyze the entire population. In one instance, we used a technique called kernel regression. It allowed us to define statistical characteristics for each customer flashcard. It gave us a number for the slope, changes in activities, frequency of particular events. To your point, looking at the pictures first gives you a idea of what variables really matter. Once we had all this descriptive data for each customer, it was easy to put customers into behavior-based buckets.

John Dillard

James,

Your points here are excellent. I am new to the blog, so forgive me if I am retreading old ground here.

I work with a mix of government and commercial clients and frequently run into snags when we start talking about which customers are most important, particularly when there are, as you describe, an awful lot of customers.

Step one is great; we too work to define segments that make sense based on available data on customer behavior. We find that basing on the customers' core business behavior, rather than on their buying/use habits with the seller/provider, is more effective.

The second step is the tough one. Because we focus on setting priorities, we have conversations with our clients about prioritizing those segments--which is a useful application of the flash cards you describe. Organizations have great difficulty in prioritizing segments when the criteria aren't always about revenue and profitability. For the service provider, there are segments that are more "important" because serving them extends the provider into a growing segment, or a segment that is tied to building capabilities that are fundamentally important to the future of the business. I intent to test out some of your ideas to see how they work in this context.

Love the column, I will keep reading.

The comments to this entry are closed.

Search Site


  • dmblog.fico.com

Subscribe

  • enter your email

Upcoming Events

  • FICO Tools & Analytics User Forum 2012
    BERLIN: September 11-12, 2012 LONDON: September 18-19, 2012 Gain new insights for improving business performance through advanced analytics and decision management tools.