Unison is Tribeca Software's analytics package, which provides the basis for a new paradigm of web intelligence infrastructure. It works seamlessly with current web architectures, integrating in a straightforward manner with a site's current web and application servers. Unison tracks not only what is requested by a visitor to the site, but what was presented to the visitor as a result of that request, and what if any business events occurred as a result of this request/response cycle.

Such a system is vital not only for measuring the effectiveness of dynamically generated promotions and advertisements, but to track and understand business events such as placing or removing an object from a shopping cart, consummating a purchase, signing up for a newsletter, clicking on an advertisement, moving from one sales department to another on the site, and so on. Only by combining such business level events with the clickstream can useful visitor-centric behavior reports be generated. And understanding visitor behavior at this level of detail permits actionable decisions to be made.

Unison consists of three components:

1) The Collection System

2) The Transformation and Warehouse System

3) The Reporting System

The following figure presents a schematic overview of the Unison architecture:

The collection system is that portion of the package that generates documents which augment the server logs and which describes the dynamic data the server sends to the browser as well as the business events which take place. Small footprint modules are installed on the web server and application server machines; these modules extend the servers' activity by way of an easy to use API in order to create XML documents which represent information about the events for which we would like to feed into the warehouse and subsequently analyze. The collection system also consists of a transport mechanism, which is at once reliable and robust, and which does not add much overhead to the existing machine processes or network bandwidth allocation. Thus the impact upon an existing site's architecture and physical resources are minimized, satisfying one of Tribeca Software's important design goals.

The XML documents generated above are collected at the receiving end of the collection system, which usually resides on a CPU separate from the application servers. Here they are assembled into larger documents, each one representing all of the events which belong to one request/response cycle. This involves correlating the pre-defined business events associated with the requested page with both the static content and the dynamically generated content of that page.

At this point the XML documents enter the transformation system. It is the task of this system to process the information contained within the documents and load them into an Oracle 8i warehouse. The structure of this warehouse is carefully designed to permit a multidimensional view of the data, and to support efficient OLAP style reporting. It is in this portion of the Unison system where set level aggregations are calculated in order to make the OLAP report generation possible in real-time. Here too data type conversions and the conforming of the dimensions of the data model are implemented. The following diagram represents this transformation:

While much of the foregoing is technical in nature, and hard to understand without a grounding in data warehousing technology, we designed Unison to effectively hide these complications from the end user. After a member of the web site's technical team has configured the system, the only visible piece of the system is the reporting system.

The reporting system is an easy to use front-end where the site's business managers and planners get to put Unison through its paces. By firing up the reporting system component and connecting to the warehouse, a wide variety of web site metrics are available for viewing and investigating. Questions about visitor sessions on the site which could not be asked in the past are now easily explored. A small sampling of the types of information which can be explored using Unison include:

• What characteristics are shared by visitors who return often? Does the time between visits predict other behavior, such as tendency to buy?

• What kinds of paths do visitors take through the site? What is the profitability of those customers who view several informational pages?

• How long does a visitor dwell on any particular page? Is this tied to other characteristics of the visitor?

• What products were sold on the site? What can we say about the visitors who bought any particular product, or product type?

• Did the visitor participate in community activities such as chat rooms or newsletters? Can we categorize the behavior of those who do?

• Why does a customer take an item out of their shopping cart or abandon the cart altogether?

• What are the characteristics of a profitable customer? What are their acquisitions costs and customer support costs?

• How can we measure the effectiveness of promotional campaigns? How many times does an advertisement need to be presented to a visitor to be effective? And within what time window?

• Were attempts to cross-sell or up-sell the customer successful? What characteristics of the visitor predict this outcome?

Note that while it is possible to use the drill down capabilities of the reporting system to examine an individual visitor session, the real power of the software comes from asking about groups of visitors. As a concrete example, we can ask whether those visitors who came to the site between five and seven times over the past month and who were exposed to an advertisement for a particular product more than three times during those visits, went on to purchase that product. This is precisely the kind of information which is needed in order to evaluate how to maximize the return on investment for promotional campaigns.