Study Measures Keyword Performance in Paid Search Advertising
New Haven, Conn., December 10, 2007 – Oliver J. Rutz of the Yale School of Management and Randolph E. Bucklin of the UCLA Anderson School of Business have developed the first model to measure the performance of individual keywords in paid search advertising on Internet search engines.
“Our model can help advertising managers evaluate which keywords in a paid search campaign are the most profitable,” said Rutz, assistant professor of marketing at Yale. “The model is also predictive and can be used to generate an optimal list of keywords.”
In paid search campaigns, advertisers bid for specific keywords to display a text ad in the sponsored section of the search results page. Advertisers are charged when a user clicks on the ad. Many keywords lead to few, if any, sales, even over several months. This sparse sales conversion data makes it difficult to measure the profit performance for individual keywords. As a result, campaign performance is typically assessed by managing large groups of keywords together or by using easy-to-calculate heuristics such as click-through rate.
Rutz and Bucklin’s model addresses the data sparseness problem by taking advantage of similarities between keywords and using a statistical method to estimate the probability that keywords with few or no sales will lead to sales conversion in the future. With the estimated conversion rate, the authors can calculate the monthly cost-per-sale of each keyword.
The authors applied the model to the keyword-level paid search data from a Google campaign for a major lodging chain. Of the 405 keywords the company bid on, 301 led to at least one click during a one-month period, and of those, 84 led to hotel reservations. Using the model, the authors estimated the sales conversion rate for each of the 217 keywords that did not lead to a reservation during the sample period and evaluate the cost-per-sale, or in this case, cost-per-reservation.
Advertisers can look to a few factors to help predict conversion rates of keywords. Rutz and Bucklin found that the position of the keyword, click-through rate, and keyword characteristics are significant predictors of conversion rates for keywords.
Rutz and Bucklin found that the model has predictive power by using it to identify a list of optimal keywords for the lodging chain. Of the 301 keywords that generated clicks in the chain’s paid search campaign, lists of the 150 most profitable were selected using the model and other common campaign management strategies, including keyword category management and click-through rate. The model’s list outperformed those of all other strategies by more than $1,200 in two months. That is an improvement of $8 per keyword over two months or $48 per keyword over a year. Click-through rate was the worst strategy for profit maximization.
“Click-through rate is a very popular management strategy in paid search,” said Rutz. “Our results show that it doesn’t maximize profits, and in the case of this study, performed even worse than just discarding all keywords that didn’t lead to reservations for the lodging chain.”
The paper “A Model of Individual Keyword Performance in Paid Search Advertising” is available online.