Monday, May 29, 2006

The Rich Get Richer: Google Needs Some Ad Sense

After 29 years of working in high-tech companies and writing about them, I have noticed how insular they tend to be, often not seeing either the world or themselves at all clearly. Whether intended or not, this cultural artifact comes to control how the world in turn sees them, which rarely works in their favor. The classic example is Microsoft, where hiring smart people fresh from school and working them 60 hours or more per week -- in an environment where they don't even leave the building to eat -- leads to a state of corporate delusion, where lying and cheating suddenly begin to make sense. But it isn't just Microsoft that does this. It is ANY high tech company that hires young people, isolates them through long hours at work, feeds them at work, and effectively determines their friends, who are their co-workers. This trend even extends to the anti-Microsoft, to Google, where the light of day is sorely needed.

Google is secretive. This started as a deliberate marketing mystique, but endures today more as a really annoying company habit. Google folks don't understand why the rest of us have a problem with this, but then Google folks aren't like you and me. The result of this secrecy and Google's "almighty algorithm" mentality is that the company makes changes -- and mistakes -- without informing its customers or even doing all that much to correct the problems. It's all just beta code, after all. But the business part is real, as is the money that some people have lost because of Google's poor communication skills combined, frankly, with poor follow-through.

First there is click fraud. Google makes its money when people click on Google ads, but some of those clicks are fraudulent -- are not honestly intended to gain information or to buy products. Click fraud generally comes in two varieties that I'll call "buy" and "sell." An example of buy-side click fraud would be my little sister religiously clicking on every Google ad on this page (What? We have no Google ads?) in the mistaken belief that doing so would make me some money. It is mass clicking by a single person without an intention to actually buy or even to gain information. Sell-side click fraud would be one advertiser clicking on the ads of a competitor with the intention of costing that competitor money without increasing their sales. Both types of click fraud ought to be detectable, and in fact, Google says it already detects the 10 percent or so of clicks that are fraudulent (Business 2.0 magazine says it is more like 30 percent), and adjusts the bill before the advertiser even knows what is happening.

But not all click fraud is detected automatically. It is one thing to notice the same IP address being used to click 30 ads in three seconds (that is obviously fraud), but quite another if the clicks are spread out or come from what appear to be a variety of users. There, too, Google pledges to make things right, though it may take some time -- too much time, I think.

My friend Mario Fantoni is a victim of click fraud, which in this case is simply defined as his Google AdWords bill climbing from $250 one month to $4,000 the next with no change in the campaign or increase in sales. Mario contacted Google, which, after an "investigation," decided that he was, indeed, a victim of click fraud. Good for Google! But that was seven weeks ago and Mario is still waiting for his credit card to be reimbursed for $3,750. Google has yet to explain why it is taking so long for Mario to get his money back. For that matter, Google has yet to actually say that Mario will GET his money back. They are still "investigating," which could mean anything because the company will not explain what it means.

If you have been a victim of acknowledged click fraud on Google (where Google admits there is a problem), please let me know about it this week (bob@cringely.com), and especially tell me what it took to get reimbursed.

The next problem I have with Google came to me courtesy of Luis Dias, a software developer with IO Software in the UK. Luis's product is a mathematical equation editor cleverly called Equations! It is great for users of Don Knuth's LaTex page formatting program, especially if they want output readable in other applications, like Microsoft Word. You can find more about the program in this week's links.

Luis decided to sell his program online using a Google ad campaign, targeting terms like "physics equations," "equation editor," and of course "LaTex." Because he didn't expect much competition selling equation editors, Luis thought that he could get most of these words for about Google's minimum price, which in the UK is 1p. In practice, though, he found that the minimum price was 3p for most words, and that minimum shortly jumped and then jumped again until some words cost as much as £2.75 (about $5.15). Since there was no competition for these ads, Luis couldn't figure out what was going on, and frankly, Google wasn't much help. They said that his words had low "Quality Scores," which meant that the minimum charge per word had to go up by the amount specified. That made no sense to Luis or to me, so I contacted Jeff Huber at Google.

To his and Google's credit, Jeff became very involved in explaining this situation and Google's position. I happen to think it is the wrong position, but at least we have had good communication.

What I learned is that the Quality Score of Luis's words was low, suggesting that it was doubtful many readers would find the ads useful or click on them. "In the recent past," explained Jeff, "ads with low Quality Scores were disabled -- i.e., not shown -- which could create frustration for an advertiser since it was a binary (on/off) decision. More recently, we evolved from the binary approach to a more flexible economic model that instead of disabling lower performing ads entirely would allow them to participate as long as the minimum bid was set at an appropriate level. This feature has both provided advertisers with greater control, as well as helped reduce the number of low quality ads by better aligning economic incentives."

This explains Luis's surreal experience of having to pay more and more to get less and less. Unfortunately Google doesn't do a very good job of explaining this change, perhaps because it appears to be precisely the kind of paid placement they do at Overture Systems (now part of Yahoo). This almost total lack of explanation may have been part of the reason why Luis was scratching his head.

We have two concepts here -- a Quality Score for words and a History for campaigns. A low Quality Score can lead to an increase in minimum word price and a poor history (lots of low Quality Scores along with low clicks-through) can lead to minimum prices being raised not just for one word, but for all words. In the case of Luis, all this took place in three days and half a dozen words, which I'd hardly call much history, yet there is very little for him to do about it short of canceling his Google account and starting all over.

The funny thing about history at Google is that it exists even before people see the ads. The AdSense algorithm takes a look at ads before they are posted and makes a quality estimate that is the starting point for the history even before an ad is shown. It apparently looks at ads in preparation, too, which is how it is able to assign minimum bid prices. Even ads you decide never to run can affect prices.

There appears to me to be a fundamental error here in Google methodology. While they talk a lot about keywords and their quality scores, the Google system appears not to work with keywords at all, but with campaigns. A poorly performing keyword will drag down all the other keywords in the campaign no matter if their quality score is good or not, if they had just two impressions, or were ever even active.

What's happening here is just that Luis is trying to sell something that hardly anybody wants to buy.

The Google system -- THOUGH I AM SURE IT IS NOT INTENDED TO OPERATE THIS WAY -- works poorly for small sellers trying to reach buyers of obscure products. This may come down, frankly, to Google's concept of small business, which I have never seen defined.

The system appears to be optimized for people selling goods of interest to millions of people, to huge marketers no matter what they are selling, but not to tiny businesspeople trying to mine narrow niches like equation editors. By targeting campaigns and not keywords, big advertisers can use any keywords they wish regardless of the relevance simply because they exist in a much larger corpus. Because keywords are treated by Google not individually but in the context of the whole campaign and these companies receive so many clicks overall, none of their keywords are likely to perform poorly using this algorithm. The result is that no matter how inappropriate or even offensive, those words will also be cheap to buy. And because these keywords have performed "well," their "history" is then updated with regard to the campaign of the big advertiser, effectively blocking out small advertisers who can't compete with the massive click rates of big companies.

Google says "we'd much rather show nothing (white space) than a poorly targeted or non-relevant ad." But on the basis of pure performance -- what Google actually DOES, rather than what Google SAYS -- it would appear that ad quality is irrelevant in the presence of huge ad budgets. The data suggest Google really cares about massive click rates, which under most circumstances come from big companies that have a huge built-in advantage.

So the rich get richer.

Google attracts advertisers like Luis with the idea that their ads will be cheaper because, frankly, they are selling something that is only thinly traded. The dream is that the system scales and scales fairly, only it isn't fair at all because if Amazon wants to advertise an equation editor USING EXACTLY THE SAME AD TEXT AND FORMATTING AS LUIS -- their words will cost 100 times less than the same words bought by Luis. It's not that Amazon (or any other big Google advertiser) has better copy writers, it is just that they sell a broader range of things.

"A large percentage of impressions & clicks do have £0.01 minimum bids," said Jeff from Google, "but these are our very highest quality ads/advertisers."

In other words, the minimum word price is 1p, BUT NOT FOR YOU.

But what is Luis to do, I mean really? All his keywords are now at very high prices and will not come down. The only way to escape this vicious circle is to open a new account, but even then he'll still be at the mercy of badly behaved keywords that come with the equation editing territory. He has to describe his product SOMEHOW.

It would be far better, I suppose, for Google customer service to simply suggest Luis not use Google ads to sell his equation editor.

I am sure this is not what Google intended, but it is misleading, unfair, and poorly explained.

"The system does scale fairly, and provides a level playing field for both small and large advertisers," says Jeff Huber. "If Mr. Dias has relevant ads, keywords, and landing page, he should be able to do just as well as other advertisers, regardless of size. It does not mean, however, that Mr. Dias or any other advertiser will be able to economically show ads that are not relevant and not consistent with user intent. If Mr. Dias or other advertisers want a large quantity of untargeted impressions, there are a variety of media that offer these relatively cost effectively (e.g., web banner ads, TV, newspapers, magazines). It is fair to observe that if there are any advertisers who may have a slight advantage, it's advertisers who have strong brands that users recognize and trust, and therefore users find more compelling when they show relevant ads -- but that's very consistent with the 'real world' and value of brands."

It all comes down to the AdWords algorithm and its intent, which isn't to help Luis OR Amazon, but to simply maximize profit for Google.

source:http://www.pbs.org/cringely/pulpit/pulpit20060525.html


Comments: Post a Comment

<< Home

This page is powered by Blogger. Isn't yours?