Skip to main content
Beat Google at Its Own Game: Apply Predictive Analytics for Keyword Bidding

(Posted by Guest Blogger, and Avid Internet Surfer, Ian Turvill.)

It's an interesting, though probably little known, fact that when sponsored keywords appear around Google's search results, which link is given most prominence is the result of a complex auction process. 

Google’s selection process for preferred links is something of a black-box.  The cost of each keyword depends on market demand, which can fluctuate, and companies are charged on a pay-per-click basis. Importantly, this revenue model leads search engines to give preferred positions to ads with a high probability of being clicked on.

Being kept in the dark about how much others have bid leaves advertisers with little choice but to engage in active experimentation. And so they commonly find themselves bidding on multiple search terms, in numerous combinations and at different price levels, and using various tricks of the trade in the hope of securing their desired position while adhering to budgetary and other constraints.  The problem, of course, is that experimentation can be both time-consuming and expensive, especially when the context for experimentation is a system that obscures some of the key causal relationships.

Proving that Fair Isaac really does "drink its own Kool-Aid", we recently applied the principles, techniques, and technologies of EDM to address this specific problem in our associated consumer brand site, where individuals can buy their bureau credit scores.

We are exposed to information about credit scores almost continuously.  Ads on the radio, TV, and Internet blast us about their importance to credit-granting decisions.  Naturally, when consumers see or hear an ad, one of the first ports of call is Google.  Fair Isaac's challenge was to make sure that the link to was among the first that consumers completing this search saw, while also making sure it did not overpay for the privilege of being listed as such.

An Fair Isaac R&D team embarked on an initiative to develop a set of analytic and software capabilities to optimize search advertising strategies for myFICO. The goal was to create a “strategy machine” that could analyze and act upon relevant data to make smarter and faster keyword bidding decisions.  The full description of the approach they took and the techniques they relied upon is presented in some detail in the most recent edition of my colleague Jeff Zabin's quarterly newsletter, Marketing Decisions.

Given that Jeff has given it such a thorough airing, I won't repeat the details here.  But, suffice it to say, the results were very impressive.  A test of the resulting strategy machine demonstrated the ability of the model to substantially increase sales volume without incurring diminishing returns. Subsequent implementation of the entire keyword portfolio in Google has demonstrated the ability to reduce keyword bidding costs by upwards of 50% while increasing ROI by 75%. For the myFICO core score portfolio, these percentages translate into millions of dollars in cost reduction per year.

related posts