As we know, purchase patterns vary tremendously across customers. Nonetheless, there are often recurring patterns for products that are purchased together. A retail store is an example of where this occurs. If these purchase patterns are understood, they can provide valuable marketing opportunities.

Product affinity analysis is a set of methodologies that looks for groups of items that tend to be purchased by the same customers. Recommendations systems (including collaborative filtering) are a related set of methodologies that can identify likely purchases at the individual level based upon past purchasing patterns. I would like to provide some insight into how these can add value to marketing strategies. But rather than running through a litany of seemingly obscure data analysis methodologies, I will lay out the marketing actions and results that they can support. If these marketing actions seem relevant for your business, then the right methodology can be determined to support your objectives.

The following section will discuss some marketing actions that can be supported using product affinity analysis.

Marketing Actions

1. Make a Recommendation

Customers are usually not simply searching for the lowest price or the best quality for a given product. At the same time, customers often suffer from a lack of knowledge about which products best suit their needs. As a result, they need to understand what products of interest are available. If the seller can provide reliable recommendations, this can tip the scale so that a customer who would have put off the buying decision now has enough information and confidence to make a purchase.

Sellers can perform this service because they know something the customer doesn’t: they know what other customers have purchased. With this data, they can

  • Look at past purchases by this customer,
  • Find other customers who have made similar purchases, and
  • Determine what additional purchases have been made by these other customers.

If the other customers tended to make well-informed decisions, then it is likely that the individual customer will also want some of these other items. (This process is made even more powerful if the seller has collected information from other customers about what they liked!)
The classic application is for e-commerce Web sites. Visit an online bookstore, put a book in the shopping cart and you will be given a list of other books that other customers have purchased with that book. This is simple but powerful. Make a series of purchases over time, and your recommendations will become more and more refined.

The aspect of this application I find most noteworthy is the degree of personalization involved. Huge amounts of data are processed in real time to provide a recommendation that is, if not unique, probably quite different from each of a hundred other random customers. Granted, some recommendations will miss the mark. But I know from personal experience that others will be appreciated and will lead to extra purchases.

2. Do Targeted Marketing

In the previous application, it is the customer that is gaining information from the analysis. In other applications, the seller gains information about customer behavior that enables improved targeting. In one form or another, the analysis determines what products tend to be purchased together. This can enable several different kinds of marketing actions:

  • Offers can be targeted to individual customers that include products that they are likely to purchase (i.e., cross-selling).
  • Products that tend to be purchased together can be placed together on the shelves of retail stores or in the pages of catalogs.
  • Promotional offers can be designed that include multiple products.

By knowing what products customers are likely to purchase, the seller can do a better job of influencing that purchase decision in their favor.
Clearly, this is closely related to "making a recommendation." However, the "recommendations" are usually made at the detailed product level (such as a specific book title). This is possible because they are delivered over the Web at little or no incremental cost. While targeted marketing actions can be personalized at this level, they will often address product categories rather than individual products. (E.g., an electronics catalog puts computers and printers on consecutive pages.)

3. Understand Customers

Often there is an underlying function or "experience" that consumers are trying to create when they buy specific products. Consumers who are trying to create the same experience often purchase similar groups of products. For example, consumers seeking a great home entertainment experience often buy big screen TVs, stereo receivers, speakers and DVD players. Not everybody will need all of these items, but often there is a pattern of affinity between products that suggests a motivation in addition to a behavior. And this, of course, can point to marketing messages and strategies that will resonate with this motivation.


In all of these applications, some thoughtful design of the analysis will help get the best results. One issue is how to integrate the analysis with the product hierarchy. Is the purchase of a DVD player different from the purchase of a DVD/VCR combination or can they be grouped together into one product? Should different brands of DVD players be kept separate or aggregated as one product? The answer depends partly on the marketing objective of the analysis but also on what the data says. If analyzing the data shows that staying with detailed products does not yield useful differences in their product affinities, then they can be combined.

Similarly, the breadth of products to include in the analysis is an issue. If I buy three books by Stephen King and two books on data mining, should my recommendation be based upon other people who bought these same five books? I would expect not. To recommend a novel it is probably sufficient to look at the novels I have purchased. The fact that I buy data mining books for professional purposes will have little to do with what novels I am likely to select. (But this could be tested!)

Database technology has made huge volumes of transactional information available for analysis. Product affinity analysis is one way to get the most out of this resource.

Register or login for access to this item and much more

All Information Management content is archived after seven days.

Community members receive:
  • All recent and archived articles
  • Conference offers and updates
  • A full menu of enewsletter options
  • Web seminars, white papers, ebooks

Don't have an account? Register for Free Unlimited Access