Search Head or Tail

Dec 8, 2005 • 1:23 pm | comments (0) by twitter | Filed Under Search Engine Strategies 2005 Chicago
 

Search Head or Tail

(FYI: This means how long a search phrase is in terms of modifiers/words. A “head” search query would be “hotel,” and a “tail” term would be “4 star hotel Dallas area.”)

Moderated by Misty Locke, Range Online Media

Kevin Lee, Dit-It.com

Surprised that people are in the audience since they are competing this session against “Meet the Crawlers.” Capturing the Tail – Going Broad. Millions of searches every day that are unique, these tail keywords & phrases are very valuable. Tail searchers usually know exactly what they want. Publishers can make more money because they meet the needs of searchers seeking these long phrases. Some tail searches could only occur once per month, or less. Many people are searching for the various obvious stuff, but once beyond that, people have their own ideas about how to express their needs/interests. So: how far out to go into that tail and make it worth it as marketers? Search behavior follows a power curve: if you remember geometry, it is a curve that never actually reaches the axis. Somewhat asymptotic, but flattens out eventually. Gives some related search distributions from “travel” down to “travel south America.” Knows that there are probably thousands more keyword permutations that can be considered. The value of a keyword is directly proportional to where it is on the curve. Campaigns and goals should line up with the profile of the searchers. Positive actions vary throughout the buying cycle.

Searchers using head keyword phrases have ambiguous desires and needs for several reasons. In addition to typed searches driving search inventory at the head, it also contains link driven traffic from directories and within the portal, as well as syndication partners. How far out to go? Aggressively distribute the tail keyword phrases that are worth it. Quite often, the bids are far lower in the tail than the head. How to set bids in tail? Kevin recommends starting fairly high, since searcher is a desired target since they know what they want. You need to make sure to have a high position to get a large percentage of the few searchers. This will help you get an opportunity to gather more data. Remember that in Google and MSN, you are competing with those people that are bidding on Broad Match. MSN has great demographic data available by keyword. At first glance, Google and MSN seem to have systems that are friendly, due to broad match ability, which “casts a wide net.” But, not all people who are doing the searches have the same intent. Specific creatives should be written within each ad group in order to increase CTR. How to find new tail opportunities? Web analytics software, Campaign management technologies and raw log files will reveal great tail keyword phrases. Tail keywords that are short benefit from the dynamic keyword insertion tool (DKI). Generally the CTR will increase with DKI, increasing efficiency. They have seen up to a 27% DKI efficiency increase.

MSN also has a DKI, but with a different structure and no ability to place a default keyword in case the phrase is too long. Yahoo Standard Match always trumps Advanced Match regardless of bid. This means for more work, requiring you to predict searches as far out in the tail as practical. When looking at the head, you probably want to segment differently. These people do not know exactly what they want. Use other parameters such as geographic, or day-parting. Hard to get enough data in the tail to day-part, but since ROI is so high in the tail you would never probably want to use day parting anyway. Between the high numbers of keywords in the tail and the high number of targeting options that make sense for the head, the data becomes significant. In the tail, competitive reactions are less frequent, meaning more elasticity in the market. When do you “kill a tail keyword?” Use statistics to “come to the rescue” cluster analysis can help. Good thing about looking at head and tail separately allows for reduction of waste, targeting of best customers, and increased profit. Improve your messages and offers, and be more aggressive when it matters.

Harrison Magun, Avenue A - Razorfish Search.

Will focus on the analyses that marketers and managers need to use to help decide which keywords/what to bid. His alternate title is “Bid down or bid up you moron!” To explain how to determine what a statistically relevant sample is, Harrison uses an example of “twins.” If there are 120 people in the room, and 6 are twins, that gives us 5%. What is the likelihood that twins are 5% of the rest of population? He shows some calculus-derived results that depict a standard bell curve that shows this probability that the incidence is between 4.5 and 5.5% is 20.5% This means if we act of the smaller sample of 120, there is an 80% chance we will make the wrong decision. How big of a sample do we need to be 90% sure that the incidence is between 4.5 and 5.5%? Answer is you need 5,044 incidences to make this estimation. To translate this into the kind of numbers to make accurate PPC management estimates, Conversion rates: 1%, need 25,000 clicks to make right decision, even at 10% conversion rate, you need 2500 clicks to make sure the conversion data is 90% accurate. So lets say you have 400 clicks and the actual conversion rate is 2%, there is a 60% chance you will make the wrong decision, Don’t waste your time on insignificant data. Sedate the screaming lunatics in your organization that are demanding changes – show them the data just described. Create accurate tests. Understand how many clicks you need for a good test. If you can’t sustain bad results, then don’t test in the first place. Spread the tests out. The idea of creating multiple keywords and campaigns and tests at a time makes it more confusing. Understand factors that impacts conversion rates.

“How can I use this sexy stats stuff?” Use Excel, no need for fancy tools. Use your own categorizations – this lets you sum up related keywords that don’t have enough clicks, and make a generalization based on business similarities, such as categories dovetailing together with respect to seasonality, for example. Use Bid Management algorithms and toolsets. This is heavy math. Add the knowledge into the four levers of search: Bidding strategy, keyword creation, messaging, business intelligence. In summation, when you look at tail, understand how accurate the data is, and take that into account when making your decisions.

Q&A Could you talk a little more about clustering? Kevin: right after I took stats test, I forgot most of those things, so I hired people with bid foreheads. (laughs) The reason he likes to start tail keywords aggressively, is because if you make a quick decision, it takes even longer to prove things wrong. When thinking about decision process, you can go conservative or aggressive. Conservative would mean you may have a 50% confidence, moving up to an aggressive stance with a 90% confidence. How to create clusters? Think about it in terms of similar intent. Requires not only a good statistical basis but also a good business basis, to make the decision. Harrison agrees…look for keywords with same attributes to cluster. Not a statistician, but knows there are regression analyses that can help with this too.

Previous story: Meet The Crawlers
 

Comments:

No comments.

blog comments powered by Disqus