Resonance Sides With Service Over Suites
In the area of marketing systems, this has led to a determined effort by the few remaining independent vendors, such as Unica and Aprimo, to expand their products' scope through acquisitions and internal development. These vendors' true competition today comes not from smaller, specialized products but the still-broader enterprise suite companies like Oracle and SAP and business intelligence experts like SAS and Teradata.
Though reliance on suites is one trend in corporate systems, an even bigger trend is the movement toward service-oriented architectures (SOAs). These involve free-floating functions, called services, that can be shared by multiple systems. Rather than maintaining their own large databases, services import the specific information needed for a particular task and return a result. In other words, they are the opposite of the big suites that are tightly integrated internally but mostly isolated from everything else. SOAs offer a way to introduce small, specialized systems within a larger, integrated structure.
You might call it a battle between suiteness and lite.
Resonance (Certona Corp., 858/586-0646, www.certona.com) illustrates the possibilities of such lightweight integration. Resonance uses neural network technology to identify linkages between behavior patterns and content selections. The most common application is to predict which items a Web site visitor is most likely to buy. This is important in itself: The company claims to increase revenue on e-commerce sites by 5 percent to 25 percent.
We'll discuss later how it makes that happen. But the point for the moment is that it accomplishes this with minimal change in the Web sites themselves. All the client needs to do is place some tags on the existing Web pages. These send real-time reports of customer behavior to the Resonance server and receive relevant product recommendations in return. The technical mechanism can be Web services or, even simpler, HTML iframes (an iframe is a region within a Web page that is controlled by an external program. It often is used to let third-party advertising services present content to visitors.).
This simplicity of deployment lets Certona sell its services on a performance basis. The company measures the difference in results between the product recommendations it serves on a client's Web page and the recommendations the client would have served otherwise. It then shares the increase in revenue. Apart from the small amount of labor required to insert the Resonance tags and send a catalog of available products (typically refreshed daily), no investment is required by the client to set up a test. Revenue splits range from 5 percent to 25 percent depending on the situation.
The technology underlying Resonance is fully self-adapting: that is, it generates recommendations without users defining business rules or building predictive models. This is an important distinction that allows quick deployment, low operating cost and fast adjustment to changes in customer behavior patterns. It is particularly suitable for businesses that offer many different products, where the cost of building rules or models for each item is prohibitive. In addition to product recommendations, the approach is suitable for document searches, suggesting related Web links, targeted online advertising and individualized outbound e-mails.
In practice, Resonance works by tracking behavior as customers move through a Web site. It watches not only what they select but where they came from, the path they followed and the time spent looking at each item. The system starts by simply observing activity during a training period. This may last from a few days to several weeks depending on site traffic. During training, Resonance builds a history of how previous customers have behaved and thus learns which items are likely to be selected under which conditions. This knowledge is converted into scoring rules that generate multi-value profiles for each visitor and each product. Profiles and scoring rules are updated continuously as new behavior occurs. All this occurs anonymously; the system is tracking Web sessions, not known individuals.
Resonance makes recommendations in real time by selecting the products whose profiles most closely match the cumulative profile of the current session. The system becomes more accurate over time as it observes more data patterns and sees less-common products being purchased.
Recommendations can be improved further by adding more information to the mix, such as profiles of registered customers and standardized product attributes. But these are optional and typically much less important than behavior during the current session. The vendor states that it begins to have a useful profile for an anonymous visitor after as few as three or four clicks.
Recommendations also can be filtered against conventional rules, such as not offering a product the customer already has purchased or recommending only products in a given category. The client sets these rules, though Certona staffers currently load them into the system. An interface to let clients enter rules for themselves is planned for future deployment.
Clients do have access to daily Web-based reports on site performance, including revenue per visit, items per order, average order size and conversions. The system also reports on visitor movements from one page to the next.
One side benefit of Resonance is that the product profiles themselves can serve as a way to classify and group related items, even if no previous categorizations exist. This can be important in applications such as organizing document or media collections.
Certona, previously N-Space Technology, has been developing its technology for several years. It began work on Resonance in 2004 and released the system in 2005. It has signed about a dozen clients since then. Resonance is offered as a hosted service only.