Utilizing Association Rules to Improve Personalization of Offers to Customers

What Association Rules are, how to build them and most importantly, how to leverage them in your marketing strategy.

The concept of Association Rules has been around since the early 1990’s and has become a core tool utilized by data miners.  The process is based on the concept of discovering regularities between products in large scale transaction data recorded by systems in retail and product fulfillment. For example, the rule  “(Potatoes, Onions)--> Hamburger”  found in the sales data of a supermarket would indicate that if a customer buys onions and potatoes together, he or she is likely to also buy hamburger meat. Such information can be used as the basis for decisions about marketing activities such as, e.g., promotional pricing or product placements.   This type of analysis is useful in developing recommendations similar to what you see when suggesting “Customers who purchased product X also purchased product Y”.

The first step it to clearly define what you what to associate.  Association rules differ from Classification rules in that there are no preset outputs.  So Association rules allow more freedom for unique combinations to be discovered.  The most typical use is to associate baskets of product purchases, but this technique has also be used to help set pricing strategy or retail product placement analysis.  Sometime the Association rules can discover some “odd” relationships.  

A famous story about association rule mining is the "beer and diaper" story. A purported survey of behavior of supermarket shoppers discovered that customers (presumably young men) who buy diapers tend also to buy beer. This anecdote became popular as an example of how unexpected association rules might be found from everyday data. There are varying opinions as to how much of the story is true.  While I don’t find diapers around many beer sections in stores, I’m sure some enterprising retailer has tested this at some point.

So what tools are able to perform this type of analysis?  The usual power house applications from SAS and SPSS are of course quite capable of crunching through volumes of data to develop these rules.  Depending upon your budget and expertise either of these tools offer solutions.  If you want to keep your tool costs down you can consider a tool I recently discovered (see this link http://www.cs.waikato.ac.nz/ml/weka/)  You will need to do a bit of reading to get up to speed, but as this is freeware, I have found the time to get up to speed more than offset by this tool’s very effective performance.

Once you discover the Association rules in your database what do you do with them?

I have found these rules provide an excellent foundation to build a cross media promotional offer engine.  I have developed some great campaigns via email where a product that the customer had recently purchased is used to re-establish the relationship and then a couple of items that showed the highest support and confidence rankings via the association rule set were offered with minimal purchase incentives.  These campaigns were set-up with test and control groups and the groups receiving the communication based on the Association rules showed a 40% lift in response as compared to the control group.

You can also create tables of product associations to feed recommendations on your website.  These can be very powerful as compared to making recommendations based on less robust analytics.

This technique can also be useful in determining how to sequence product presentations in your catalogs and brochures.  Just as retailers utilize this analysis to help layout merchandise in stores to be more consistent with customer shopping patterns, this analysis helps do the same for printed materials.  Typically the improved sequencing of products in printed material improves the response by 5 to 10 percent.

All these actions result in improved sales, retention rates and ultimately profits. 

How much of your Website Business is incremental versus channel shift from your offline efforts: A simple analytical answer along with a couple of methods to measure the quality of web buyers.

Almost every multi-channel marketer is struggling with the question, “how much of my web business is incremental”?  This is a key question, for marketers who have observed that as the web channel grows, typically the other traditional offline channels fall.  So how can a company determine if the web sales are purely a channel shift, or might it be incremental?

My experience is to look at what portion of your web business is coming from customer who are new to your file.  Obviously, if your web business is dominated by customers who have been on your file prior to the web activity, then the web channel activity is a pure shifting of business.  But that is most likely what you won’t find.  There will be a proportion that is new to file and the key is to compare that portion of new with your other traditional offline channels.

The table below illustrates this concept well.

New to File Proportion of Business by Channel
		Existing Buyer Business	New to File Business	Proportion existing Business	Proportion New to File Business
Offline Channels		$300	$100	75%	25%
Web Channels (Ver 1)		$25	$10	71%	29%
Web Channels (Ver 2)		$25	$20	56%	44%
Web Channels (Ver 3)		$25	$100	20%	80%

If your proportions looks like those in Ver. 1, your web business is clearly a channel shift only and should not be viewed as incremental at all.

If your proportions looks like Ver. 2, your web business is partially incremental, but your offline channels are still your primary growth source. 

If your proportions look like Ver. 3, you should view your web business as incremental, and just like any marketer, you need to understand what do these new customers cost to acquire, and is that an appropriate investment.  This brings up the next topic, how to measure quality of the web buyers.

Two of my previous articles will help with this task.  First is to run the “Buyer Matrix” on your web buyers to understand at what level do they return to make subsequent purchases, and how do those compare to other similar aged customers.  These behavior metrics will quickly help you determine if the investment made to acquire these customers was appropriate or not.  It will also provide you a relative measure of quality between your customers acquired from offline efforts versus online.  You should also review my previous article on Acquisition Cost to utilize the appropriate financial perspective on the appropriate investment returns.

Another useful tool is the Generic Scoring model.  In a similar fashion, you should compare the scores of your new to file buyers across channels.  There are many myths that web buyers are of lower quality of the offline buyer, and thus do not warrant the same level of investment.  My experience is that this is not always the case.  Be sure you understand how your customers behave using their actual purchase data.