Intelligence Through Meta Data
In my last article, Using Hierarchy in Data, I wrote about the importance of hierarchical data and how it often is neglected due to the challenges in brings. My argument presupposes that such data has been captured and formatted in a way that preserves such information. Moreover, there are opportunities to enhance the existing data with additional attributes that strengthen our understanding of customer behavior.
Here is a simple thought-problem that illustrates the point:
Your web site's home page devotes about a third of its real estate to
three featured products, which are presented as part of merchandising
agreements with manufacturers. At any given time, you have about a
dozen of these in "inventory" and your web server offers these up in
rotation. You track clicks for each featured product and can trace the
clickstream all the way through to purchase.
So far so good. But can you actually retrieve the three offers that each customer received on the home page? There is much evidence that the context in which an offer is made – in this case the assortment of three items – can have a significant influence on the acceptance of the offer. Could you systematically identify which combinations of offers generated the highest interest and purchase rates? Most companies discard this data, and retain only the actual log file data (alas, some companies fail to track even this information). Admittedly, on large sites, the amount of data generated would be quite large, but here is information that is central to generating revenue.
Let's take the example one step further: do you have sufficient meta data about these featured products to make meaningful decisions about what to feature in the future? Because merchandising arrangements and SKUs change often, even if you retained the combinations of offers made, the information may be too specific to be useful over time. Rather, in addition to the product-specific information (e.g. price, color, etc.), it may be important to maintain some additional meta data. For instance, it may be that products that boast the newest features (but that are more expensive) do best when offered in combination with a product with a standard set of features, but also perform poorly when offered with an obsolete, but very inexpensive, item.
Such information would also yield valuable insights on a customer level. For example, what if some customers always buy the newest products, regardless of what other items are offered in combination? That suggests opportunities for constructing a more tailored set of "featured" products to those customers that seem to be technology-oriented. If such offers can be constructed from the merchandising inventory, then the retailer, manufacturer, and customer may all benefit from more relevant offers.
In this example, we could create a meta data element that describes the product as "new," "mainstream," and "obsolete." And here, we probably need to keep some time-series history, as today's leading edge product will be obsolete some day.
I recognize that creating such meta data represents a challenge, both from the perspective of developing appropriate attributes, as well as creating and managing tags. For a site that may have several hundred thousand SKUs, it may not be practical to do this for every product. On the other hand, there could be tremendous value in doing this just for the featured products, since their importance is elevated due to the merchandising dollars supporting their placement and (presumably) the higher purchase rates that promotion generates.
Here are a few additional areas, in which these concepts can be applied:
- Catalogs: Could you recreate the essence of a catalog that you produced six months ago? [Retrieving a PDF of the catalog does not count.]
- Weekly circulars: For many retailers, circulars are an important marketing vehicle, particularly since they are supported by merchandising. Similar to the web and catalog examples: do you have the ability to reconstruct the circular from stored information and meta data?
- Digital marketing: Emails, PURLS, and digitally printed mailers may feature variable content and offers. Can these be recreated to reflect accurately what each customer received?
- Content sites: Can you identify what types of information users are seeking? In other words, do you have meta data that categorizes content so that you identify patterns that may cut across some of your content areas (for instance, I may not really be interested in Sports or Entertainment, but I will read content in these sections when there is some scandal involved).
The full list of applications would encompass many more areas. Stated more broadly, the real goal is to create a complete picture of the company's interactions with its customers across all channels. And in order to accomplish that, we need to retain and store both the operational data and enrich it with contextual information that generates more meaningful insights.
Feel free to share any experiences or success stories you have around this topic in the comments section.
Posted at 05:44AM Sep 01, 2009
by David King in Data |
