• by Ben Perry , December 1, 2006

This may sound heretical coming from a guy in my line of work, but automated bidding tools are sometimes bad for your paid search campaign. Many folks at SEM firms--myself included--spend a great deal of time banging the drum about how much you can gain from using technology to bid, and how much efficiency can be gained from using a good bidding agent. However, experience has taught me that there are situations when a bidding agent simply isn't the right tool for the job. Now that I'm older, grayer, and somewhat wiser, I know that the reason for this has to do with how sophisticated bidding systems work.

A colleague of mine is fond of saying that "bidding tools are not magic conversion machines." This is a cheeky way of telling people that bidding tools are machines that work on the principles of statistics. Unfortunately, people want to believe that a sophisticated bidding tool can cure a bad business model or improve an atrocious site conversion rate. They can't. Bidding tools don't write business plans or design Web pages, but understanding what they can do will help you make the most of their capabilities, which can be very impressive. It should be noted that I'm talking in general terms about sophisticated bidding tools that conduct portfolio-style optimization, not simple rules-based tools. Now, crank up Weird Al Yankovich's latest hit "White and Nerdy" while I explain the stats at play here.

Sophisticated bidding tools work on the principle of using data from the past to model and forecast what is expected to happen in the future. The most interesting everyday analogy for this that I've heard--and what passes for "interesting" here is enough to bore a librarian to tears--is those noise-canceling headphones everyone uses on airplanes. Believe it or not, they actually work on the same principle as sophisticated bidding agents, but... they also suffer the same shortcomings. In short, the headphones analyze sound patterns and introduce sounds to counter the noise frequencies that they detect. When you turn them on, their ability to drown out the sound of the jet engines is borderline miraculous. However, when the captain comes on the loudspeaker or the kid in front of you starts crying, those amazing headphones suddenly become ridiculously expensive ear muffs. Why? Because they couldn't possibly predict those noises and counter them.

Bidding tools can have the same basic shortcoming. Because they use the past to predict the future, any unpredictable event may cause problems. Consider the campaign for a retailer of sports merchandise when the Boston Red Sox won the World Series. The campaign manager--who happens to be a huge Sox fan--had the presence of mind to alter the settings on the tool in order to capture the increased demand that the tool couldn't possibly have predicted, especially since the last time the Sox won the series there weren't any televisions around, never mind computers or the Internet. Clearly, this is where having a little human intervention trumps technology.

Bidding tools often have another problem that noise-canceling headphones don't, which is a lack of data. The headphones only need to have data for the past few minutes, or perhaps seconds, to predict the next few minutes. Some bidding tools have the ability to predict seasonality over the course of a year and react accordingly. But this means that the tool needs to have a year's worth of data and a sufficient quantity of it--one or two conversions a week just won't cut it.

All this adds up to a situation wherein companies whose business plans rely on unpredictable events or companies with exceptionally rare conversion events are usually poor candidates for sophisticated bid optimization. Campaigns like that can use rules-based bidding tools, but primarily rely on vigilant human intervention to achieve success in the pay-per-click marketplace. Consider the case of a company that needed to capture the demand on searches generated by unfortunate events such as hurricanes and earthquakes. While these events happen regularly, no one can reliably predict them, and the impact they have on Internet searches is highly variable depending on the location, magnitude of the disaster and media coverage of the event, among other things. Bidding tools that lack a "meteorological optimization" feature (i.e., all of them) don't fare well on campaigns like that.

The good news is that most Web sites are able to take advantage of sophisticated bidding tools if they measure well and retain granular data on their campaigns. Understanding what these tools are attempting to do is key to making the most of their capabilities. Sophisticated technology has its advantages, but having clever humans running it is what will allow you to make the most of your investment in paid search technology.