• by Laurie Sullivan , Staff Writer @lauriesullivan, June 17, 2009

Amadesa has a behavioral targeting platform that learns over time without manual input from anyone. The software-as-a-service (SaaS) platform, based on machine learning technology, runs on a predictive modeling engine that identifies categories, or microsegments, that score and select the most relevant content to serve up relevant ads. The engine focuses on ecommerce sites.

The company has three patents pending. Two are related to the BT algorithm, which Avi Kedmi, Amadesa cofounder and CTO, says personalizes each visit to a Web site. The engine took two years to build. It moved into beta last year and officially launched this week. The ads are served through Akamai.

The engine assigns two sets of scores: one to the content, another to the person who visits the site. Everything occurs in real time, as the individual hits the Web page. It doesn't allow customers to set rules in the BT engine. Customers can't go into an interface and create rules to modify the behavior. The user uploads the content and promotions, banners, and images. It's done through JavaScript tagging, which Shlomo Lahav, Amadesa chief scientist, says makes the engine "true machine learning automation."

Amadesa's site-side BT algorithm, an "anonymous personalization engine," incorporates "principles more commonly seen in advanced advertising applications and [applies] them on marketers' sites," according to Rita Brogley, the company's CEO. Once the click takes the visitor from the publisher's Web site to the advertiser's site, Amadesa begins collecting hundreds of data points to determine the most relevant piece of content for the visitors when they hit the Web page.

Site visitors see different content on the site based on the hundreds of attributes collected. Those include time of day, day of week, browser, operating system, domain, and other variables contained in the URL. The engine also knows the keywords that brought the person to the site.

Roughly 10 Amadesa customers currently run behavioral targeting across their Web sites, according to Pete Olson, Amadesa vice president of product management. For example, Smooth Fitness, a fitness equipment retailer, garnered online traffic comprised of many unique sub-sets. Unfortunately, the company had no way to tell which specific topics were attracting that traffic, so it served up the same promotional message to all visitors.

In the first month after Smooth Fitness began relying on Amadesa's behavioral targeting platform, the company achieved a 12.6% click-through rate lift, as compared with the random serving of content, according to Olson. During the second month, BT drove a 48.3% lift. By the third month, the click-through rate rose to 123.4%. Smooth Fitness is on track to generate an incremental $1.3 million of additional revenue, according to Amadesa.

There are challenges. Olson admits companies still have concerns about privacy and personalization of content. "They want to know how to serve up more relevant ads without getting intrusive," he says.

Retargeting and reporting are on Amadesa's product road map, but not offered today. The company also will build an application programming interface (API) that allows other providers to tie into the algorithms. These providers might have an internal ad serving site. The product would launch by the end of the year.