By Karen Burka

PPC advertisers have a range of choices when it comes to keyword optimization and bidding approaches, from rules-based to local and global models-based solutions. But only one of these approaches – global keyword-level modeling – can lift performance from 25% to 200%, particularly for large, complex PPC campaigns.

There are two main approaches to bid optimization: rules-based and models-based. A models-based system uses historical performance data to train statistical models to predict future performance. For example, a models-based system could predict the bids necessary to achieve a 200% return on ad spend (ROAS) tomorrow. Rules-based systems, in contrast, use a pre-defined set of reactions to certain situations, i.e., "if ROAS is less than 200%, then lower bids by 10%."

Many PPC advertisers mistakenly believe that models-based bid optimization software is a black-box technology that is only appropriate for tail terms, and that head terms should be manually optimized. This misperception is based on some of the early marketplace solutions, which lacked transparency and typically delivered only modest short-term performance gains that tapered off over time.

Today’s bid optimization technology has developed to the point that it can maximize financial performance across all keywords while being transparent and meeting critical business needs. For PPC advertisers managing a large number of keywords, some form of models-based bid optimization technique is virtually essential.

The Benefits of Global vs. Local Optimization

Local and global optimization are two distinct models-based approaches. Local optimization bids each keyword in a PPC program separately. For example, if the target ROAS is 200% then every keyword is bid to obtain an individual ROAS of 200%. A local solution won’t trade off a low ROAS on one keyword for a high ROAS from another.

Global optimization (also called a portfolio approach by some vendors) considers all keywords at once, assigning bids so that, on average, the group as a whole maximizes a goal while meeting certain constraints, such as a specific ROAS or cost per action (CPA). The advantage of global optimization is that it treats each keyword appropriately with respect to the whole campaign.

One keyword can drive significant revenue at 180% ROAS while another drives the same amount of revenue at 220% ROAS. As long as the average ROAS is 200%, the global solution is successful. This approach generally provides higher value from a set of keywords than local optimization.

Global Keyword-level vs. Cluster-based Modeling

To drill down even further, PPC advertisers can choose either keyword-level or cluster-based modeling within global optimization. Clustering aggregates data from a few keywords to several hundred or thousand keywords, which are assumed to have the same general performance characteristics. The advantage to using clusters is model stability, meaning the results are repeatable. Clustering was developed to assemble data sets large enough to permit analysts to apply more traditional statistical techniques to determine bids.

But there are three substantial drawbacks to cluster-based modeling.

• Individual keywords lose their uniqueness in a cluster. While statisticians might value the extra data, individual keywords lose their uniqueness in a cluster. The result is a loss of performance compared to modeling each keyword as a unique entity as the advertiser spends too much on some keywords in a cluster and too little on others.
• Cluster models need to be manually tuned. Clustering typically requires statisticians to tune the models manually. As a result, cluster-based solutions are rarely pure software applications and can’t offer the degree of automated bid optimization needed by advertisers managing thousands or millions of keywords in a dynamic bid-based marketplace.
• Cluster models periodically need to be retrained. All statistical models decay over time, making it necessary to periodically retrain the statistical models involved. As a result, cluster models tend to be used far beyond their lifecycle with a resulting negative impact on performance.

Global keyword-level optimization treats each keyword individually and regularly improves performance, including profit, revenue and ROAS, by 25% or more in controlled tests against global cluster-level, local keyword-level and rules-based technologies. Keyword-level modeling can be done on keywords with as few as 10 conversions a year.

In addition to the financial benefit stated above here are some additional benefits:

• Automation – Only pure software can accomplish global-keyword level optimization.
• Transparency – because it works at the most granular level, global-keyword optimization offers visibility into the performance variables, models, bidding decisions and results.
• Control – marketplaces are dynamic and changes sometimes need to be made quickly. Global-keyword optimization reacts immediately to changes in goals and/or constraints.

Utilizing Appropriate Math and Transparency

To maximize effectiveness, PPC advertisers need the right balance of appropriate math and software automation, as well as transparency into the specific variables that drive individual keyword performance.

PPC advertisers do not need to cluster to create sufficient click or impression data when they use a mathematical approach to global keyword-level modeling. While clustering might produce an acceptable "average" for all keywords in the cluster, it has little relevance to the future performance of the individual keywords contained in the cluster. Modeling the performance of individual keywords within the cluster – even those with small amounts of data – can be done by using the appropriate mathematical modeling techniques. These techniques include structural risk minimization, a technique that trains models to become simpler as data sets become sparser, as opposed to complex clustering techniques that often build models that are far more elaborate than the data will support.

For large PPC advertisers that want to avoid sacrificing long-tail keyword performance, automation is usually an advantage in a dynamic advertising marketplace, both in terms of lower cost and superior daily bid optimization of all keywords. The same type of software automation techniques can be applied to retraining the models that predict keyword performance.

Transparency is the ability to see the individual variables that drive keyword performance. Modeling keywords individually allows advertisers to make more intelligent determinations on each and every keyword in a PPC campaign by viewing the decisions the software made and why.

Successfully using bid optimization software requires understanding the advantages and limitations of the different models-based techniques and the recognition that there are no set-and-forget solutions. Keyword-level, global bid optimization is the ultimate solution for deriving maximum profit from PPC campaigns that involve a large number of keywords. It’s the right choice for optimizing head terms, tail terms and everything in between.

Karen Burka is a freelance writer specializing in online marketing and advertising topics.

About This e-Solution Spotlight…
e-Solutions Spotlights are produced in association with sponsors by Third Door Media, publisher of Search Engine Land. The opinions expressed are not those of the Search Engine Land editorial staff. Interested in being an e-Solution Spotlight sponsor? Contact us by filling out this form.