yieldbuild_logo.jpgYieldBuild aims to tweak the appearance of contextual ads to maximize revenue.

Earlier this week, I wrote that its technology was less sophisticated than that of a similar company called Rubicon Project, which we covered here — and did so based on available reports on the company. After meeting with YieldBuild’s founder and watching a demo, I’ve changed my mind

Placement and formatting have a dramatic effect on the performance of contextual ads. However, optimization involves significant manual effort and, worse, a set-up that works for a while often starts to slip. This may be because site visitors get accustomed to the ads and stop noticing them.

This is where YieldBuild comes in. You give it a set of “zones” for ads, input the maximum size of these zones, and let it go to work. Its algorithms automatically gauge your ads’ performance and make constant adjustments to their color, placement, and positioning. This is different than Rubicon, which focuses on the selection of ad networks but does not manipulate placement and format. Pubmatic, a similar company, aims to optimize network selection and offers a simple interface to change the ads’ style. In its beta phase, Pubmatic does not automatically optimize all formatting aspects of the ads, but these features are in the pipeline, and the market will only become more competitive.

The problem with all of these services is that pricing is the most important factor, and real time analysis of pricing is not possible: Some networks report CPMs two weeks after the fact. Optimization will remain a somewhat clumsy process.

The adjustments that YieldBuild makes can be large: It may choose not to use a zone at all, or place an ad in a small part of it. It may make the background black and the text a light gray, or match the page’s color and make the ads blue. The adjustments can also be subtle, indenting the text of the ads by a few pixels or making the background just a tiny bit lighter than the other elements on the page.

It is also responsive to the source of the traffic, so someone who navigates to a customer’s site directly will see a different arrangement than someone coming from a search engine.

It sounds arbitrary, but it’s not. As Malcolm Gladwell’s book Blink suggests, our conscious decisions are influenced by the most subtle of cues, so a slight difference in text color or position can have significant effect. Coming from a Google search, having just seen results and ads displayed a certain way, for example, you might be more likely to click on ads that match the web site’s page. As YieldBuild’s founder, Paul Edmondson, says “we don’t know why it works, we just know that it does.”

Paul claims that once the system learns the formatting and layout that works best on your site, which takes around 100,000 page views, it boosts revenue by between 32 percent and 100 percent, with an average of 47 percent. The major changes are made during those first 100,000 views, so sites with lots of traffic will optimize quickly, while smaller sites will take much longer. He says that a future release will learn with around 25,000 page views.

YieldBuild faces the risk that a site will use it until it has made the big changes and then turn it off. Paul says that keeping it running will ensure that the ads’ performance won’t slip over time.

YieldBuild’s customers are not all small fish. Though Paul could not go on record about them, we discussed a few, and more than a handful are notable. He says he has a number of large companies waiting to sign up, but doesn’t yet have the capacity to take them on.

The San Francisco company takes three percent of the revenue from impressions on its customers’ sites. It has raised $2 million from Hummer Winblad Venture Partners.techcrunch_screenshot.jpg