Individual-keyword level modeling delivers superior results
Individual keywords vs. clustering: At the most granular level, there are two ways to train models around data for paid-search bid optimization: keyword and clustering. There are significant differences between the two that have a direct impact on PPC campaign performance and management.
In keyword modeling, every keyword is treated as an individual market where the performance data for each is considered, independent of the others. Models are trained using the individual data for every keyword, leading to bids for each keyword that are as unique as the data it generates.
Because each is treated individually, keyword models can be retrained frequently and, because all models decay over time, the ability to retrain easily is critical to ongoing optimized performance. In addition, the expansion and reduction of keywords is easy to accomplish, making PPC campaign management easier and more efficient.
Clustering puts similar-sounding keywords into groups, aggregates the data from each and determines one bid that is assigned to each member of the cluster. The purpose is to get around the problem of sparse data, but in doing so, clustering ignores the unique data generated by each keyword and, therefore, cannot “see” what drives performance.
Clustering will result in increased performance at the beginning, but as the models decay, performance will suffer. Retraining clustered models is time- and money-intensive because the new performance data for each member of the cluster has to be gathered and aggregated before the model can be built, whereas keyword model systems can continually update performance data.
Similarly, the expansion or reduction of keywords in cluster modeling is not as easy as keyword modeling. Doing so requires breaking and re-clustering, which, like retraining models, is a time- and money-intensive project.