Secrets of Google Electronics, Part 2
This is a continuation of the article " Secrets of Google Electronics ".
Self-focusing on engineering points, mathematical formulas and data mining made Google a completely new type of company. But in order to fully understand why, you should go back a little and look under the roof of AdWords.
Most people think that Google’s ad auction is a very “straightforward” tool. But in fact, there is another key point that not only a minority of users of the system itself knows about, but even not all experienced advertisers are aware of its existence. Bids themselves are only part of the system that determines the auction winner. Another, perhaps the most important factor determining the winner of the auction is the so-called. “Quality score” (the quality score). This value was created in order to determine whether the advertisement that users are showing is exactly suitable for its description and requirements for it, i.e. roughly speaking defines the "truth" of an advertising message. For if not, then the entire system suffers and the company ultimately receives less money.
')
Quality Score is calculated on the basis of many other indicators, including the relevance of advertising keywords, the quality of the page to which the advertising link leads, and, above all, the percentage of unique clicks on the ad unit when it appears on the final page (obviously, there are other factors but Google will never discuss them for obvious reasons). In addition, there is a mechanism for "punishing" advertisers in the event that the quality of advertising is below the average level - in such cases, the company automatically assigns the minimum rate to the advertiser. Google explains that this tactic, brought to the public by many companies and individuals who are automatically “underestimated”, protects users from irrelevant or annoying ads that discredit the name of contextual advertising or paid links in general. Already even managed to go through several court cases from individuals who stated that they were victims of arbitrariness in a quasi-monopoly.
And if it is possible to argue about the fairness of the algorithm, then it can be called a “arbitrariness” with a stretch. In order to set a “quality indicator”, a company sometimes has to be ahead of the curve to count how many users click on an advertisement. This is a very delicate moment, especially when we are talking about billions of mini-auctions. But since the entire model depends on the accuracy of the prediction of clicks, Google should as best as possible collect and process each "curl" of information. Susan Wojcicki, who oversees the entire advertising business of a company, calls this "click physics."
When Hal Varian spent his second summer in Mountain View, he still came only once or twice a week. It was then that another turning point occurred - Mr. Varian asked the recently hired computer specialist from Stanford, Diane Tang to create an analogue of CPI (Consumer Price Index) for the company, which was called the Keyword Price Index (KPI). “Instead of a basket with comrades like beer and donuts, we have keywords in it,” says Tang, who is known inside the company under the pseudonym Queen of Clicks.
KPI is a kind of test for reality and compliance with reality. This indicator warns Google if abnormal “price bubbles” appear in the system - a sure sign that the auction does not work as it should. Categories are determined by the cost per click, which the advertiser usually has to pay, and are ranked according to their distribution, and then they are divided into three branches: upper, middle and lower. “The top line is a very popular word, like“ hotel ”or“ flowers, ”Tang explains.“ In reality, the middle branch has words that can be seasonal, for example, snowboard is unlikely to be popular at the end of the season. Well, the lower branch is rarely used keywords and phrases, like "Chelyabinsk Metallurgical Plant". "
The index invented by Diana Tang is a good example of how, with a large amount of information, you can create effective mechanisms for improving and regulating the system itself. In addition, it is obvious that such tools must exist and function in order for AdWords to work and make money.
“The people who work with me, for the most part, are a mixture between economists and those who are interested in statistics,” says Varian, who in 2007 finally moved to Google for a permanent job, and now he leads two groups, one of which is closely involved in analyzing .
“Google needs mathematical formulas and algorithms to create tools to find signals in noise” - these are the words of one of the statisticians Daryl Pregibon, who joined the company in 2003 after 23 works as lead scientists at Bell Labs and AT & T Labs. “Our rule is simple - one statistician per 100 computer scientists.”
Keywords and clickthrough rates are bread and butter systems. “We try to understand our own mechanisms not in the form of numbers and indicators, but in general, as a living organism,” notes Qing Wu, one of Varian's henchmen. Wu specializes in forecasts, so he has to predict how the business situation will change depending on the season, climate, international holidays, even the time of day. "We have information about temperature, weather and queries, so we can perform statistical modeling and adjust this result during various changes." All these results are delivered in a digestible form to the other end of the system - the advertiser, which allows you to more effectively manage your own advertising.
In order to track and test forecasts, Wu and his colleagues use a variety of different tables that receive information in real time. A kind of terminal Bloomberg for Googlesphere. Wu constantly checks the results in order to compare the accuracy of prediction with reality: “This is a handy tool that allows you not only to see the wrong side of requests, but also how much money and advertisers we have, which keywords are popular at the current time, and what is the level of” return "for each of the advertisers."
Wu calls the company "the barometer of the whole world." And in fact, to study the clicks in some way similar to the situation when you could look out into an imaginary window from which the entire planet is visible. You can follow the change of seasons - the clicks react to a change in the direction of skiing and warm clothing in the cold season, at the same time showing how popular bikinis are in summer. We also see, for example, the one who is most popular among representatives of pop culture. Many of us remember some key events from newspapers and TVs, but on Google, all this information is displayed in the form of peaks of various graphs. “One of the most serious ups occurred several years ago during the SARS epidemic, ” says Tang. But King Wu did not even have to read the newspaper in order to learn about the financial crisis - he saw an extraordinary leap in search queries on the word "gold." From all this it is possible to understand how important are the forecasts and analysis of information, every drop of which, as we see, matters.
After Google hired Varian, many other companies, such as Yahoo, also decided that they needed a head of the economic department who would oversee the team involved in a thorough work related to auctions, advertising and economic models in order to lick their business to shine. In 2007, another Harvard economist, Susan Athey, was surprised to receive a personal meeting with Steve Ballmer in Redmond. “This is not just a call, it’s a challenge that you take,” recalls Susan, who had worked in Microsoft offices in Cambridge and Massachusetts in previous years.
So, it turns out, the rest of the world is left somewhere behind? Eric Schmidt believes that if this happens, it’s not as fast as many people think. He also believes that with the help of auctions, similar to the one that Google invented, you can conduct any type of transaction. What is the solution to the problem of oversaturation of the automotive market? Put the whole range of unsold cars up for auction - it will be cleaned out in record fast time. The same situation with the real estate market: “Now people go to the auction in case of dissatisfaction, when there are practically no other options to sell anything. But imagine a situation in which such an approach would be normal and quite natural for the majority, ”says Schmidt.
Hal Varian believes that the new era is not far off, which he himself calls the “information statement” - its essence is in mastering supply and demand. “What's convenient and cheap?” Varian asks, and immediately answers: “Information.” So what is missing to make it all happen now? Analytical abilities for mass analysis of this very information. Hal Varian believes that there will come a time when a person with a technical background or a propensity for engineering approaches will work in hedge funds on Wall Street and make bold decisions based on analyzing information and statistical data, while obtaining unique results.
Perhaps this approach is indeed satisfactory for Hal Varian, who does not hide the fact that he chose an economist’s career inspired by science fiction: "In the first book of Isaac Asimov from the Foundation trilogy, there was a character who built the mathematical models of society and it seemed to me that it really entertaining. When I went to college, I looked everywhere for myself that would allow me to do the same. It turned out to be the economy. ” Hal Varian tells this story while sitting at his ranch, where he sometimes stays in order to avoid the 40-mile journey from the company's headquarters to his house. Now this ranch belongs to Google - it was here that Brin and Page started the business of the company.
The wild contrast between how Spartan furnished the house and what happened as a result of its existence - the emergence on the planet of dozens of geek millionaires, as well as new rules for doing business in the information world, makes an impression. And, perhaps, this contrast is much more surprising than what Azimov wrote about 60 years ago. What could be more surprising than the largest company operating in the capitalist world and, at the same time, giving its best services to anyone who wants it for free? Varian, of course, does not suffer from the complex of God and is well aware that he owes success not because of inspirational insanity, but to an early definition of the potential of the Internet and concentration on speed, volume, analysis, statistics and customer satisfaction. A pinch of auction theory does not interfere with the general taste. And today there is a word to describe the whole dish as a whole: Googlemonics. It will have to learn, or pay the price and still trail at the end.
Source of English text: Wired Magazine
Self-focusing on engineering points, mathematical formulas and data mining made Google a completely new type of company. But in order to fully understand why, you should go back a little and look under the roof of AdWords.
Most people think that Google’s ad auction is a very “straightforward” tool. But in fact, there is another key point that not only a minority of users of the system itself knows about, but even not all experienced advertisers are aware of its existence. Bids themselves are only part of the system that determines the auction winner. Another, perhaps the most important factor determining the winner of the auction is the so-called. “Quality score” (the quality score). This value was created in order to determine whether the advertisement that users are showing is exactly suitable for its description and requirements for it, i.e. roughly speaking defines the "truth" of an advertising message. For if not, then the entire system suffers and the company ultimately receives less money.
')
Quality Score is calculated on the basis of many other indicators, including the relevance of advertising keywords, the quality of the page to which the advertising link leads, and, above all, the percentage of unique clicks on the ad unit when it appears on the final page (obviously, there are other factors but Google will never discuss them for obvious reasons). In addition, there is a mechanism for "punishing" advertisers in the event that the quality of advertising is below the average level - in such cases, the company automatically assigns the minimum rate to the advertiser. Google explains that this tactic, brought to the public by many companies and individuals who are automatically “underestimated”, protects users from irrelevant or annoying ads that discredit the name of contextual advertising or paid links in general. Already even managed to go through several court cases from individuals who stated that they were victims of arbitrariness in a quasi-monopoly.
And if it is possible to argue about the fairness of the algorithm, then it can be called a “arbitrariness” with a stretch. In order to set a “quality indicator”, a company sometimes has to be ahead of the curve to count how many users click on an advertisement. This is a very delicate moment, especially when we are talking about billions of mini-auctions. But since the entire model depends on the accuracy of the prediction of clicks, Google should as best as possible collect and process each "curl" of information. Susan Wojcicki, who oversees the entire advertising business of a company, calls this "click physics."
When Hal Varian spent his second summer in Mountain View, he still came only once or twice a week. It was then that another turning point occurred - Mr. Varian asked the recently hired computer specialist from Stanford, Diane Tang to create an analogue of CPI (Consumer Price Index) for the company, which was called the Keyword Price Index (KPI). “Instead of a basket with comrades like beer and donuts, we have keywords in it,” says Tang, who is known inside the company under the pseudonym Queen of Clicks.
KPI is a kind of test for reality and compliance with reality. This indicator warns Google if abnormal “price bubbles” appear in the system - a sure sign that the auction does not work as it should. Categories are determined by the cost per click, which the advertiser usually has to pay, and are ranked according to their distribution, and then they are divided into three branches: upper, middle and lower. “The top line is a very popular word, like“ hotel ”or“ flowers, ”Tang explains.“ In reality, the middle branch has words that can be seasonal, for example, snowboard is unlikely to be popular at the end of the season. Well, the lower branch is rarely used keywords and phrases, like "Chelyabinsk Metallurgical Plant". "
The index invented by Diana Tang is a good example of how, with a large amount of information, you can create effective mechanisms for improving and regulating the system itself. In addition, it is obvious that such tools must exist and function in order for AdWords to work and make money.
“The people who work with me, for the most part, are a mixture between economists and those who are interested in statistics,” says Varian, who in 2007 finally moved to Google for a permanent job, and now he leads two groups, one of which is closely involved in analyzing .
“Google needs mathematical formulas and algorithms to create tools to find signals in noise” - these are the words of one of the statisticians Daryl Pregibon, who joined the company in 2003 after 23 works as lead scientists at Bell Labs and AT & T Labs. “Our rule is simple - one statistician per 100 computer scientists.”
Keywords and clickthrough rates are bread and butter systems. “We try to understand our own mechanisms not in the form of numbers and indicators, but in general, as a living organism,” notes Qing Wu, one of Varian's henchmen. Wu specializes in forecasts, so he has to predict how the business situation will change depending on the season, climate, international holidays, even the time of day. "We have information about temperature, weather and queries, so we can perform statistical modeling and adjust this result during various changes." All these results are delivered in a digestible form to the other end of the system - the advertiser, which allows you to more effectively manage your own advertising.
In order to track and test forecasts, Wu and his colleagues use a variety of different tables that receive information in real time. A kind of terminal Bloomberg for Googlesphere. Wu constantly checks the results in order to compare the accuracy of prediction with reality: “This is a handy tool that allows you not only to see the wrong side of requests, but also how much money and advertisers we have, which keywords are popular at the current time, and what is the level of” return "for each of the advertisers."
Wu calls the company "the barometer of the whole world." And in fact, to study the clicks in some way similar to the situation when you could look out into an imaginary window from which the entire planet is visible. You can follow the change of seasons - the clicks react to a change in the direction of skiing and warm clothing in the cold season, at the same time showing how popular bikinis are in summer. We also see, for example, the one who is most popular among representatives of pop culture. Many of us remember some key events from newspapers and TVs, but on Google, all this information is displayed in the form of peaks of various graphs. “One of the most serious ups occurred several years ago during the SARS epidemic, ” says Tang. But King Wu did not even have to read the newspaper in order to learn about the financial crisis - he saw an extraordinary leap in search queries on the word "gold." From all this it is possible to understand how important are the forecasts and analysis of information, every drop of which, as we see, matters.
After Google hired Varian, many other companies, such as Yahoo, also decided that they needed a head of the economic department who would oversee the team involved in a thorough work related to auctions, advertising and economic models in order to lick their business to shine. In 2007, another Harvard economist, Susan Athey, was surprised to receive a personal meeting with Steve Ballmer in Redmond. “This is not just a call, it’s a challenge that you take,” recalls Susan, who had worked in Microsoft offices in Cambridge and Massachusetts in previous years.
So, it turns out, the rest of the world is left somewhere behind? Eric Schmidt believes that if this happens, it’s not as fast as many people think. He also believes that with the help of auctions, similar to the one that Google invented, you can conduct any type of transaction. What is the solution to the problem of oversaturation of the automotive market? Put the whole range of unsold cars up for auction - it will be cleaned out in record fast time. The same situation with the real estate market: “Now people go to the auction in case of dissatisfaction, when there are practically no other options to sell anything. But imagine a situation in which such an approach would be normal and quite natural for the majority, ”says Schmidt.
Hal Varian believes that the new era is not far off, which he himself calls the “information statement” - its essence is in mastering supply and demand. “What's convenient and cheap?” Varian asks, and immediately answers: “Information.” So what is missing to make it all happen now? Analytical abilities for mass analysis of this very information. Hal Varian believes that there will come a time when a person with a technical background or a propensity for engineering approaches will work in hedge funds on Wall Street and make bold decisions based on analyzing information and statistical data, while obtaining unique results.
Perhaps this approach is indeed satisfactory for Hal Varian, who does not hide the fact that he chose an economist’s career inspired by science fiction: "In the first book of Isaac Asimov from the Foundation trilogy, there was a character who built the mathematical models of society and it seemed to me that it really entertaining. When I went to college, I looked everywhere for myself that would allow me to do the same. It turned out to be the economy. ” Hal Varian tells this story while sitting at his ranch, where he sometimes stays in order to avoid the 40-mile journey from the company's headquarters to his house. Now this ranch belongs to Google - it was here that Brin and Page started the business of the company.
The wild contrast between how Spartan furnished the house and what happened as a result of its existence - the emergence on the planet of dozens of geek millionaires, as well as new rules for doing business in the information world, makes an impression. And, perhaps, this contrast is much more surprising than what Azimov wrote about 60 years ago. What could be more surprising than the largest company operating in the capitalist world and, at the same time, giving its best services to anyone who wants it for free? Varian, of course, does not suffer from the complex of God and is well aware that he owes success not because of inspirational insanity, but to an early definition of the potential of the Internet and concentration on speed, volume, analysis, statistics and customer satisfaction. A pinch of auction theory does not interfere with the general taste. And today there is a word to describe the whole dish as a whole: Googlemonics. It will have to learn, or pay the price and still trail at the end.
Source of English text: Wired Magazine
Source: https://habr.com/ru/post/63579/
All Articles