There are many variables that I like to analyze, on an annual basis, in an Online / Retail Dynamics project.
Merchandise Categories: I sum annual demand by merchandise category, then divide the total by how much the customer spent in the past twelve months. This gives me a fraction (0.00 to 1.00) of amount spent in each category. Some clients want two years or five years or all history included. I find this is not an optimal way to analyze what customers purchase - who cares that you purchased a love seat in 2004? In these cases, I weight historical spend ... maybe 100% for 12-month purchases, 50% for 13-24 month, 30% for 25-36 month, 20% for 37-48 month, 14% for 49-60 month, and 10% for 61+ month purchases. This greatly minimizes the influence of old purchases, especially older high-dollar purchases.
Website Visits: Here's a little secret not many folks want you to know - in most of my projects, I'm asked to analyze twelve-month website visitation behavior. Having said that, on average, only website visits in the past 15 - 30 days have any influence on future behavior. Often, I'll create three variables ... website / mobile app interactions in past 30 days, then from 31-90 days ago, and finally, 91-365 days ago. But again, only the most recent website / mobile app interaction matters. Recency is critically important online, folks. Heck, sometimes I'm asked to group website visits into buckets ... 1 visit last year, 2 visits last year, 3-5 visits last year, 6-10 visits last year, 11-50 visits last year, 51-100 visits last year, 100+, that kind of thing. Whatever works for you is fine, just make sure you have a defensible point of view.
Purchases: I like to sum twelve-month (or historical weighted) dollars by channel ... retail, smartphone, tablet, desktop/laptop, call center, that kind of thing. Then I'll divide the totals within channel by total annual (or historical weighted) dollars.
Website Characteristics: Here, I like to categorize activity on a weighted basis ... 100% for 0-30 day activity, 20% for 31-90 day activity, 5% for 91-365 day activity. I'll create 1/0 indicators for all key characteristics (cart, email click-through, referral from Bing, that kind of thing), then I will weight each characteristic by time (100%, 20%, 5% as mentioned above), and create a percentage. The weighting becomes important ... if a customer visited via Bing 100 days ago and Google yesterday, the Google visit is weighted at 100%, the Bing visit at 20%, meaning that the customer has a Google preference at a rate of 100/120 = 83%, while the customer prefers Bing at 20/120 = 17%. On an annual basis, the weightings really help us understand how the customer behaves.
Store Distance: I'll plug 1/0 indicators into my analysis for 0-5 mile bands, 6-10 mile, 11-25 mile, 26-50 mile, and 51+ mile bands. You will learn that visitation behavior changes as customers get further and further away from a store.
Zip Codes: I categorize zip codes by Catalog-Centric, Online-Centric, and Retail-Centric. Behavior in each classification is simply different, and quite interesting! You probably have your own algorithm for categorizing each zip code, so use that.
Tomorrow, I'll show you how I cook this information up - the discussion might get a bit geeky, but that's the nature of the work I'm doing when analyzing Online / Retail Dynamics.
A few days ago I told you a story from my time at Eddie Bauer. As the new Circulation Director, it didn't take long for the problems ...
Look at the first four rows of our life table (values of 0/1/2/3). These are the first 12-15 weeks after a customer buys for the firs...
You probably run Life Tables for your customer file, right? Right? They've been around forever ( click here for a reference f...
If you don't like geeky math, please skip this post, because I am about to show you how the sausage is made! I have eight variables in...