Advanced Paid Search Techniques
Moderated by Danny Sullivan, Editor of Search Engine Land
First speaker will be Eduardo Llach from SearchRev. He will be talking about various advanced PPC techniques. He will assume that the crowd already knows the basics. He introduces some variables for optimization of campaigns. Many factors influence user behavior and success.
First he talks about multivariate targeting. For the term “online dating,” the conversion rates are higher and lower based on geographic location, so you can adjust campaign accordingly. If you have a particular area that has already been identified as being “hot,” you can go deeper and use syndication to target more searchers. You can then add the time of day and week during which to focus efforts. Lastly, the creative for the ads can be changed based on all of the previous.
Geo /metro targeting. This uses the user’s IP address which is included in the HTML header, for Google, Yahoo!, and MSN servers to see. Country mapping is very accurate, but State mapping is accurate only to 50%. He shows a spreadsheet which outlines some different metro areas, and compares CTR, average pos, conversion rate, average CPC and the CPO. The chart is intended to show the significant deltas in the various metrics across various metro areas. They have found that grouping together the top 10 metro areas allows the campaign to perform as well as a national campaign, but with a far greater number of conversions.
Syndication is useful too, and this can be measured across portals on which the ads are delivered. He shows another chart which breaks categorical areas into auto motive, retail and financial services, across the Google system (Google, Search network, and Content network).
How to structure a campaign that works? Optimize beyond the keyword level, using multivariate optimization. Then he briefly discusses day parting. Rules: track results each day, focus on conversion rate and CPO, bid according to the Conversion rate (if conversion rate goes up 20% on a Tuesday, bid up 20% on Tuesdays). He shows an example which was 28% more effective through the addition of day parting.
Jon Kelly is next up, from Sure Hits. His topic will be long tail keywords and geo-targeting. What is the “keyword long tail?” “Home loan” is a “head” term that generates a tremendous amount of traffic, versus thousands of phrases that generate only one search per month or year, such as “Topeka Kansas home equity loan rates.” These small volume terms add up. People who use these terms are indicating a clear intention. This can yield a better conversion rate. Also, since less people are bidding on the long tail terms, this can lead to lower CPC.
3 action steps that they take at Sure Hits: 1. calculating click value. For a low-volume phrase, discover the probability of conversion and the value of conversion. In this case, there are 3 attributes that most people use: geography, product, and type of request. The wrong way to analyze this data is in buckets. Do not use all Oklahoma in one bucket, all home equity loan terms in another bucket, etc. He advocates the use of “tagging keywords,” almost like tagging pictures or blog posts. Tags help to predict conversions. For example searches containing “city” and “rates” are lower converting phrases, while a state-inclusive search tends to be a higher converting word. Also look at value of head versus tail. For home loan, the click value may work out to $7, where the longer tail term could be worth $15. Again, this is modeled using the tags.
2nd action step: reward the users’ choices. Use market model to tag the keywords, and then serve up a specific landing page which offers the consumer exactly what they are looking for. If you go the extra mile and systematically show the consumer what they are looking for, you will have better results.
3rd area: watch campaign data. One of the reasons that city phrases are often problematic, is that they are also brand phrases. A search for “Phoenix mortgages” could actually be for the Phoenix mortgage company, and a click on your listing will be a waste of that is the specific brand they are already seeking.
Watch out for “fake tail phrases,” homonyms (Mobile home loans instead of “Mobile Alabama home loans” leads some to think mobile home and pothers to think home in Mobile, AL). They found through research that some geo targeting was simply not working, like in the case with Houston Texas. They removed Houston, and it made a huge difference in the campaign. Looking for areas of a campaign where something doesn’t look right, and acting on that through research, is key.
Matt Van Wagner from Find Me Faster, presenting “New Fish, Old Socks, and a New Attitude.” He will talk about dynamic keyword insertion, or DKI. He has heard lots of different comments/feelings about this. Has heard that it improves CTR and quality score, that it requires a secret sauce, and that it improves ad relevance. These are all good myths. Cons include that you lose control of what the ad will look like, it can be complex to understand, and it can have an impact of decrease in conversion.
Showed some nightmares, including a search for “used fish” at Google, then shows “used cigars,” “used underwear” (lots of laughs). So what is DKI? At its core, it is a way to customize ads and help improve relevancy. This transforms generic ads into customized ads, according to MSN. Matt hen shows an example of “Starbucks coffee.” The keywords inserted come from the keyword list, not from what the person typed into the search box, as some people think. He then shows a cool way to capitalize on word casing by changing the order of capital and lower case letters within the individual word (cool stuff).
Think twice before inserting Google dynamic text in the display URL. Matt warns that you have to be careful with broad matching. Use “red sneakers” in your keyword list, not “sneakers red,” or that is the way it will come out. With expanded broad matching these can be issues as well. He shows the used underwear ad again at Google, and how it comes up with an ad for “suede thongs.” That business is actually trying to sell the shoe version of thongs, but it has been included in the results due to expanded broad match’s semantic mapping of “underwear” to “thong.”
Matt then briefly discusses the Yahoo! Panama version, and he seems to mostly like it. He goes through the different ways to specify alternate and default texts to use instead of the DKI term triggered by the search. He says that MS “went all out” on dynamic text. He is very impressed, it seems, by their system. “They give you a lot of functionality.” He shows a cross platform and checks with capabilities that each has, and only MS has checks next to each capability (Matt is starting to speed up and I could not catch them, but it was “Title, description, …”)
Last speaker will be Michael Sack from Idearc media. He will discuss day parting, Hi-Lo optimization, and portfolios. Day parting - when to use: when have competitive market conditions; limited sales windows; identifiable demographics by time; business capabilities; having detected patterns in conversion/ROI. Prerequisites are pretty stringent. You have to have good analytics to do it, and you should do it on an hourly basis, he feels, in order to be most effective. You have to be able to plot performance versus time. Of course, you also have to have the ability to adjust keyword on an hourly basis.
He shows a spreadsheet with obvious spikes in conversions, which dictate when the bids would want to be raised or diminished. Idearc has a pretty cool day parting setup interface, as well as a rules-based system he will discuss shortly.
What is Hi-Lo optimization? Comes right out of day parting. Business all have a cycles, doing better at some times of the day than others. Capitalizing on the “good times” are the “Hi periods.” This theory of analysis suggests most phrase shave a high performance window. He tested this theory by checking for consistency between three different businesses in the same industry and none of them had similar High times. This indicated that it is based on the businesses individual customers, and not the industry category.
Keywords are to be treated like investments. They have a cost, an expected return, there is a risk, you can track performance, and understand and monitor market behaviors.
One problem: bidding is a bit like playing black jack. “The dealer must hit on 16…” But if someone else on the table is “uneducated” they may hit when the dealer is showing six, and give the dealer a chance to win even though the other players held. In the example he shows, “Joe” hits on a 14, gets a nine, and then the dealer draws a 5 on the 16, this beating the other two players due only to Joe’s mistake. This is very much like some of the idiots that run Paid Search campaigns, using “ego bidding” or other things that mess up the dynamics of the particular keyword universe.
He says to treat portfolios like investments, set rules and objectives, and measure performance. Diversify the portfolio by using more keywords and leveraging the long tail. Then, bid more effectively by day parting, and mine data for patterns for the best Bid Time Zone or BTZ (trademarked). Portfolios work with good realistic rules and a large segment of keywords with them to populate.
(This is live coverage of SES San Jose 2007, and some typos or grammatical errors may exist. If you were a panelist and you would like something clarified, please post in the comments or contact me through the system)
