ADOTAS – Keyword bidding techniques are not created equal. With four incredibly different types of approaches, understanding their differences is a challenge in and of itself. But knowing the differences and applying the correct approach to your paid search program and campaign goals can translate into significant performance gains.
Now the question is when and how to use the choices available to you.
Over the past few years, digital marketing platforms and consultants have helped advertisers optimize paid search spend using a combination of human analysts and rules and model-based bid optimization software. In this case, technology has played a major role in scaling the challenge of daily bidding on tens of thousands to millions of keywords, from the head to the extreme tail. However, there is still confusion around when to use rules-based bidding or the three major model-based bidding techniques: local optimization, global cluster modeling and global keyword-level modeling. We’ll explore the differences between each approach and examine the outcomes.
Rules-based bidding is all about reacting to situations, rather than predicting and adjusting by learning from historical conversion data. Advertisers who forego using tail terms and instead focus only on head terms often use rules-based bidding. This strategy works when you’re only dealing with a few keywords and have the resources to leverage human analysis, versus optimization software. Of course, this method can only scale so far, as it requires paid search professionals to provide analysis and crunch the numbers on the fly.
With model-based approaches, it’s important to make the distinction between local and global optimization. Local optimization bids each keyword so it achieves the goal in an SEM program, while global optimization (sometimes referred to as a portfolio approach) considers all keywords at once. Using the global approach, bids are assigned so that, on average, the whole group is maximized for a goal. The global approach usually provides higher value from a set of keywords, compared to local optimization, because with local, some keywords are overbid, while others are underbid to achieve the goal, and that results in poor financial performance.
Global Optimization: Cluster-Modeling
Within global keyword modeling there are two approaches: cluster-based modeling and keyword-level modeling. Cluster-based modeling was developed to cope with the data problem surrounding tail terms. Because there was very little historic data for these terms, clusters aggregated data from hundreds or even thousands of keywords, allowing analysts to apply traditional statistical techniques to determine bids. Keyword clusters lead to prediction stability — however, each keyword is actually unique, which ultimately results in a loss of performance when compared to modeling each keyword individually.
Clustering has another consideration – the manual tuning needed to optimize the keyword clusters means human analysis is needed on a regular basis on top of the software required to automate the bidding for the clusters. Also, it’s very time-intensive to create the clusters to begin with. In fact, this is the biggest issue, because as a cluster ages, performance decays. Combatting this becomes very expensive.
Global Optimization: Keyword-Level Modeling
As mentioned before, keyword-level modeling is another option for advertisers — however, many believed it couldn’t be done, because there wasn’t enough tail term behavior data available to build accurate models. This is no longer the case. Keyword-level modeling is possible, due to optimization technology developments and sophisticated mathematical techniques.
Leveraging advanced math, software automation and transparency into specific variables that drive individual keyword performance, advertisers can create models for all keywords in an SEM program, not just the low-hanging fruit of head terms. This can result in overall performance gains of over 25 percent compared to clustering, local optimization and rules-based techniques.
For advertisers looking to create only brand awareness, rules-based bidding works toward driving traffic. Rules-based bidding is a good option for advertisers who aren’t seeking revenue — rather, they’re working toward impressions.
Meanwhile, advertisers who aren’t operating with constraints can leverage local optimization. If there’s an explicit goal — for example, to drive as many orders as possible, without regard to cost or profit — local optimization will achieve those results. Granted, most advertisers do have constraints, which makes global clusters an alternative if advertisers only need to track conversions at the campaign level.
For advertisers who want more transparency and deal with large numbers of keywords, keyword-level modeling provides the opportunity to automate bidding while avoiding the performance and maintenance issues associated with clustering. Because this type of modeling provides transparency into the variables that drive keyword performance, keyword-level modeling can result in dramatic campaign gains when dealing with tens of thousands to millions of keywords, whether they’re head or tail terms. Compared to the other keyword bidding techniques, keyword-level modeling is ideal for advertisers looking to maximize both head and tail terms, while having transparence and control over a SEM program.