The following is the fourth in a series of excerpts from “Bid Testing Best Practices”, a white paper from OptiMine. In it author Jason Mulvey explores the three primary methods for conducting bid testing in paid search. In earlier posts we covered position-based bid testing and current bid-based bid testing. In this post, Jason looks at value-based bid testing and how it differs from its two cousins. Visit optimine.com to download the whitepaper.
Value-Based Bid Testing
Value-based bid testing differs significantly from its two cousins. Whereas position-based and current bid-based approaches each test around a point that has no relation to actual keyword performance, value-based bid testing uses keyword performance as its guide. Regardless of position or bid, value-based testing seeks to test around the observed value of a keyword.
Properly implemented, value-based bid testing reduces the risk of loss while maximizing the probability of improved performance. In practice, value-based testing works by moving up and down around a bid that matches the keyword’s average value per click. For example: If a keyword receives 100 clicks and returns $100 in revenue, the average value per click is $1. Value-based bid testing will use that average value as a starting point, testing up and down in a reasonable range and monitoring keyword performance.
Where value-based testing becomes difficult is in estimating the value per click for every keyword. Not every keyword receives a click and, of those that do, there is always a percentage that does not experience a conversion event. In cases where there is no conversion or value data, predictive modeling is required to estimate the value. While difficult, estimating the value per click can provide immediate improvements to performance, especially in face of changing goals and external conditions.
By nature, the appropriate predictive models used for bid testing are exploratory. Traditional predictive models do not provide estimates outside the range of values that have been seen in the past. Therefore, if a keyword has never had a bid as high as $2, a predictive model should not predict a $2 bid. Exploratory models have that restriction removed, allowing them to test in new ranges of bids that have not been tried before. The disadvantage to this exploration, however, is that the predictive accuracy of the performance is greatly decreased because there is no keyword performance data for the bids being tested.
Next up, Jason will look at value-based bid testing in practice and how the same data can result is very different results depending on the goal. If you don’t want to wait, you can have the white paper in its entirety by downloading it today.