David Baker would be the first to admit he has a weird job. As Google’s director of engineering, one of his duties is to police the ads run on Google’s sites and its third-party network of publishers. That involves shutting down ads, which could seem counterintuitive for a company that banks more than 90 percent of its revenue from advertising. But as far as Baker sees it, removing ads that drive users to scam sites (among other types of bad ads) “allows us to make an honest buck.” Adweek spoke with Google’s ad cop about what it takes to patrol online advertising’s underbelly.
Adweek: You’re known as one of Google’s ad cops. What exactly do you do?
Ad cop is a nice description of a lot of what I do. I have a few responsibilities, but one is to lead a number of different engineering teams in the fight against bad ads. We firmly believe that it’s in our long-term best interests to make sure that all ads are trustworthy and that we are showing the right ads to our end-users. If users begin to distrust the ads they are shown, then regardless of whether Google makes money in the short term on that ad, Google is harmed rightly in the long term. So we work hard to build systems to detect and identify different content of different types that range from things that are illegal to things that we think are appropriate for only certain audiences and make sure that we get as many automated decisions as possible but involve real people in making the final call in questionable cases.
Not to get too in the weeds, but how do you go about evaluating the ads?
We approach this problem from three separate dimensions. We analyze individual ads. We look at an entire site in trying to determine, “Is this entire site unacceptable for some policy reason?” For example, a site that’s selling counterfeit goods, it doesn’t matter what particular product a user ends up on—if they’re selling counterfeit goods, we don’t want to do business with them. And we look at the advertiser account perspective. When we get a new account, we inspect it and try to determine if this a legitimate advertiser and how behavior for an advertiser changes over time.
How sophisticated are the people trying to game your system in order to run bad ads, and how do you keep up with them?
There are some fairly sophisticated bad actors. An advertiser who tries to slip into our system after he’s already been suspended by doing some very low-level advertising that looks perfectly fine and then later will add into their account the ads for counterfeit goods and try to get their ads served as much as they can—that’s a behavior we have seen frequently. In that particular case, we can’t analyze an advertiser only once or only right after the account’s been created. We have to continually analyze advertiser accounts and catch them when their behavior changes. A bad ad that gets an hour or two worth of visibility, that’s a huge problem.
Of the bad ads you catch, how often is it a scam like one that directs to a counterfeit commerce site versus one like an alcohol ad served to an underage user?
Every two weeks we give what we call “human evaluators” that are paid by Google but are not Google employees a sample of sites that have served weighted by the number of impressions, and we ask them, “Are these good sites or bad sites?” From 2010 to 2011, we drove that metric down by 50 percent, yet we have drastically increased the volume of accounts that [Google needs to evaluate]. The better we get and the faster we get at shutting down bad things, the harder the bad actors try to work to get back into the system. So we put them into a tighter loop. In 2011, we ended up shutting down over 800,000 advertiser accounts, and 150,000 were for counterfeiting.
You pointed out the negative implications of letting a bad ad sneak through, but is there a counter-danger of filtering out a good ad as a bad ad?
We tend to err on the side of caution, and we are going to mistakenly catch good ads and good advertisers and either delay their serving or prevent them altogether. I hope our advertisers understand the need for mistakes like that. If we delay a good advertiser from serving so that we take a close look at their account, yes, their ads are delayed from serving. However, they will be shown alongside other good ads. It harms a good advertiser to have their ads served alongside bad ads.
What do you do when a good ad gets filtered out?
Whenever we make a negative decision against an advertiser, we provide an appeals mechanism. Based on appeals, we overturned less than 3 percent in 2011. By and large, the decisions we make are very accurate, but it is important to us to have a very efficient and effective appeals mechanism.
Does mobile present any unique challenges in detecting bad ads?
Sometimes we have to build some mechanisms that are specific to mobile sites. But we view these as one big space of online advertising.
Do you guys forecast ahead a year to anticipate new developments to prepare for, or is it just honing what you have in place?
There are lots of ways we can do better: catching more bad ads, catching more bad ads quickly, catching fewer good advertisers mistakenly. It’s a constant battle with these bad actors as they try new types of scams that we may be unaware of initially or they try to find holes in our system that they don’t think we’re monitoring. I guess the nice thing for me is, I see long-term employment.