Website Morphing :: News

Executive Summary:

The authors define website morphing as changing “the website automatically by matching characteristics to customers’ cognitive styles”. They define a cognitive style as “a person’s preferred way of gathering, processing, and evaluating information”. They note four technical challenges, namely, “(1) For first-time visitors, a website must morph based on relatively few clicks’ otherwise, the customer sees little benefit. (2) Even if we knew a customer’s cognitive style, the website must learn which characteristics are best for which customers (in terms of sales or profit). (3) To be practical, a system needs prior distributions on parameters. (4) Implementation requires a real-time working system (and the inherently difficult web programming.” They proceed to solve these problems using their preferred mathematical solution.

Of particular interest is the classification of clicks based on area of the page: content area, content purpose, and navigation – and the use of this data in real-time.

They also launch into a case study by British Telecom (BT) and the math behind it, namely Gittins and Bayesian loops, concluding “The Bayesian updating enables customers to reveal their cognitive styles through their clickstreams. Together, the Gittins and Bayesian loops automate morphing (after a priming study).” The authors estimate that if their method were implemented system wide for BT, the increases in conversion would “represent approximately $80 million in additional revenue”.

In sum, they describe a process for automatically modifying the look and feel of a website based on performance and cognitive styles.

Review:

The paper is enormously important for web analytics practitioners for at least four reasons.

The first is that it provides an example of web analytics being used in real-time to improve the user experience. Incremental improvements to a website can be very challenging to execute in many medium to large organizations. The ability to set meta-meta rules for experiences and to incrementally adjust those rules based on the analytics opens a new dimension of applicability for web analytics.

The second is that finally, perhaps, the Internet Channel can live up to some of the initial 1995 hype. Supposedly, we were supposed to be able to do what Direct Mail marketers have been able to do for decades – careful segmentation and experiential differentiation. So far, many (if not most) websites, serve up one-size-fits-all experiences, or, we differentiate using very early forms of personalization. The underlining theme of morphing is to adjust the experience to the manner in which a user thinks.

The third is that web analytics, as it is currently practiced, does not rely on many statistical methods. Not all variants of website morphing will make use of the authors preferred course of Gittins plus Bayesian loops. Web analytics, as a practice, will face the dual pressures of having to acquire a new skill set and simultaneously explain the methods to non-practitioners.

Initially, I believed that website morphing would pose a direct threat to hosted vendors. While retaining hosted analytics solutions in addition to a morphing analytical system might be psychologically unsatisfying from a management perspective (“why can’t one system do it all?”), page view level analytics, even without morphing information embedded therein, can still be of some value to the organization. It can holistically report how, in the past, the website has performed.

Morphing represents an additional layer of data - a complication. Many tools do not take into consideration the effect of an A/B or a multivariate test over a specific period of time even within the same vendor toolset, so the deficiency and complication is already present. If there are 16 cognitive profiles and 16 matching morphs, with a single A/B test therein, there would be 32 unique experiences to measure. Third party analytics could tell the organization how the website morph is performing in aggregate, but significant changes to the existing pageview paradigm would need to be made for the third party provider to offer data with respect to why which morph, and which version of the morph, performed better.

Fourth, website morphing will require an initial set of inputs that will be heavily, though not exclusively, informed by web analytics derived insights. The preparation of these customer profiles and the pre-populating of experiences will represent an enormous opportunity for web analysts to shine in an interdisciplinary process. Web analysts risk a loss of leadership if they neglect the opportunity or do not seek a major ownership stake in the technology.

In sum, the paper is important, and I would recommend a read by web analytics practitioners and managers.