Hal Var­ian is the Chief Econ­o­mist at Google, engaged pri­mar­ily in the design of the company’s ‘adver­tis­ing auc­tions’; the auc­tions that hap­pen every time a search takes place in order to deter­mine the adver­tis­ing that appears on the results page.

After intro­duc­ing us to this con­cept, Steven Levy looks at Google’s “across-the-board empha­sis on engi­neer­ing, math­e­mat­i­cal for­mu­las, and data-mining” and how these ‘Google-style auc­tions’ are applic­a­ble to all sorts of appli­ca­tions.

You can argue about [AdWords’] fair­ness, but arbi­trary it ain’t. To fig­ure out the qual­ity score, Google needs to esti­mate in advance how many users will click on an ad. That’s very tricky, espe­cially since we’re talk­ing about bil­lions of auc­tions. But since the ad model depends on pre­dict­ing click­throughs as per­fectly as pos­si­ble, the com­pany must quan­tify and ana­lyze every twist and turn of the data. Susan Woj­ci­cki, who over­sees Google’s adver­tis­ing, refers to it as “the physics of clicks.”

[…] “Google needs math­e­mat­i­cal types that have a rich tool set for look­ing for sig­nals in noise,” says sta­tis­ti­cian Daryl Preg­i­bon, who joined Google in 2003 after 23 years as a top sci­en­tist at Bell Labs and AT&T Labs. “The rough rule of thumb is one sta­tis­ti­cian for every 100 com­puter scientists.”

Key­words and click rates are their bread and but­ter. “We are try­ing to under­stand the mech­a­nisms behind the met­rics,” says Qing Wu, one of Varian’s min­ions. His spe­cialty is fore­cast­ing, so now he pre­dicts pat­terns of queries based on the sea­son, the cli­mate, inter­na­tional hol­i­days, even the time of day. “We have tem­per­a­ture data, weather data, and queries data, so we can do cor­re­la­tion and sta­tis­ti­cal mod­el­ing,” Wu says. The results all feed into Google’s back­end sys­tem, help­ing adver­tis­ers devise more-efficient campaigns.