I work for a PPC bid management software company. Therefore, it should not be surprising that I advocate the use of bid management software. Bid management tools provide considerable value to PPC analysts, and I expect that most of the industry would agree with this to some extent. But somewhere along the path of promoting automated bid management, a perception has evolved that the algorithms powering bid management systems are capable of something akin to sorcery. Many SEM professionals now seem to have the idea and, even worse, the expectation that merely connecting a PPC campaign to these magic algorithms will result in dramatic and immediate performance improvements.
Was it something I said that led to this misconception? Was I overly exuberant in my advocacy of automated bid management tools? If so, I apologize and will now attempt to remedy the miscommunication with this confession of sorts.
Let me begin by stating loud and clear. . . No bidding algorithm will magically improve campaign performance in all situations. Not my company’s software, and not any of our competitors. If anyone tells you differently, they are either talking beyond their experience level or flat out lying.
Now that I’ve gotten that out of the way, allow me to explain a few reasons why the magic cure-all algorithm is, and will likely remain, a myth. First of all, we need to acknowledge that algorithms are not magic. They are mathematical processes leveraging statistical analysis of historical performance data to predict future performance. Did that sentence provide any more clarity than the magic theory? Probably not, so let me put it in simpler terms. If you flipped a coin a few thousand times and tabulated the results, it would almost certainly show close to 50% heads and 50% tails. You could then use this historical coin flipping data to predict how often the coin will land on heads and tails in future flips. We’re working with many more variables in our keyword bid management, but this is essentially the same approach a bid algorithm takes when determining an optimal bid.
An algorithm thinks a little more like this. For optimizing to a CPA target, it analyzes the historical conversion rate of a keyword to predict the future conversion rate. Once the prediction is made, an optimal bid level can then be calculated to achieve a CPA goal. An oversimplified explanation to say the least. But in very general terms, this is the way an algorithm thinks.
But here’s the rub. . . we set bids on keywords, not actual search phrases. And we all know that these do not necessarily match. Set a keyword to a broad match type, and Google will take quite a bit of latitude with the searches it gets matched to. So while it would be reasonable to assume that people looking for ‘widget repair services’ in the future will complete my lead form at a conversion rate similar to those who searched for ‘widget repair services’ in the past, we cannot blindly assume that our ‘widget repair services’ keyword will reflect that same behavior. What if Google starts matching this keyword to people searching for ‘widget delivery services’ or just ‘widgets’? Both our past and future conversion rates will become more a product of keyword/search term matching than user behavior.
I sincerely hope that readers are nodding their heads in understanding now. User behavior is tied to searches. But bidding algorithms are only looking at keyword performance, which may or may not reflect search term performance. The worse your matching is, the less effective the bidding algorithm will be. Almost every time I’m asked to review a client’s bid management setup, I find keyword/search term matching problems. I then have to explain that the bidding algorithms won’t be effective if keywords aren’t being matched to closely related search terms. If the jar labelled flour is filled with salt, you’re going to end up with some lousy tasting cookies. This is pretty much the same thing that is happening with a bid algorithm optimizing keywords with poor search term matching.
You might wonder why the algorithms don’t take this matching into consideration and adjust accordingly. Well, think about what would be necessary for a bidding tool to do this. You can’t improve matching without making structural changes to your campaigns. Is your bid management system automatically changing keyword match types and adding negatives for you? Would you even want it to do such a thing? I certainly wouldn’t. When our keyword gets matched to ‘widget mending services’, would you want the bidding system to automatically make ‘mending’ a negative? Absolutely not! Maybe someday we’ll get to a point where bidding systems have the natural language analysis capabilities to make sound structural decisions, but we’re a long ways off from it today.
Keyword/Search term matching is just one of many factors that can inhibit the success of bidding algorithms. The biggest challenge of any bidding algorithm is managing bids on low volume keywords that don’t have enough historical performance data to make accurate predictions. Take a look at your campaigns and you’ll likely find that this is the majority of your keywords. That’s normal. While these keywords individually drive a small amount of volume, they can make up a large portion of your volume in aggregate. And since they are typically the most specific keywords, they often have the best conversion rates. Let’s agree that these long-tail keywords are important.
If a keyword has only received four or five clicks in the last 30 days, how would either a person or an algorithm make a good prediction of how that keyword will convert in the future and calculate a bid? We really have only one option. Look at a grouping of similar keywords and consider how the group converts in aggregate. And this begs the question, how will we define the group? What makes keywords similar?
The metric we want to use to measure similarity is again conversion rate. But this presents somewhat of a catch22 situation. How do we measure the future conversion rate correlation between multiple keywords without a large enough click sample to accurately predict the conversion rate of each keyword independently? The process would require an analysis of individual words (often called tokens) within search phrases that across all keywords show a high correlation with a shift in conversion rates. Intuitively we are already aware of such tokens. You surely have noticed that any keyword containing your brand name tends to convert better than keywords without. Other tokens may include terms like ‘price’ or ‘quote’. Again, we are back to a natural language analysis challenge. There is a way to perform this analysis programmatically, but it is complex and not being attempted by any bid management tool I’m aware of. If anyone know of a tool that does, please find a way to let me know! I’d be beyond excited to review such a tool.
Every bid management tool I’m aware of either leaves the grouping decision to the user or bases it on campaign structure. Both of these options are less than optimal and can result in poor bidding decisions. If the bidding algorithm is grouping low volume keywords based on campaign structure and brand terms are littered throughout all ad groups and campaigns, we can expect the algorithm to under-predict the future conversion rate of any low-volume brand keyword. Just an example, but you can see how structural grouping can cause problems. Leaving the grouping decision to the user isn’t much better, as many will end up simply guessing.
Let me get back to the original purpose of this article. . . to dispel the myth that a one-size-fits-all algorithm will unfailingly improve the performance of any campaign. Both the keyword matching and keyword grouping discussions above are pointing out one critical fact. Bid management isn’t the only factor impacting campaign performance. If other areas of campaign management are being handled poorly, an algorithm could make performance worse. Note that I said “could” make performance worse. There is one more huge variable we have yet to throw into this discussion. The most important factor determining if campaigns will perform better or worse is how well bids were being managed prior to plugging in the bidding algorithm. If you put a monkey in front of a 10-key and let him determine keyword bids, you’re certain to see a massive improvement after switching to a bid management tool. No offense to monkeys. I often think a monkey would do a better job than many PPC analysts I’ve encountered. But you get the idea.
Here is a phrase that I’ve heard countless times from countless SEM software sales reps. “Companies using our bidding technology have seen an average performance improvement of 30% to 40%.” I honestly feel that these claims are what led to the magic algorithm myth, and it’s these sales people that should be making a confession. . . but sadly they probably don’t know any better. In reality, this big improvement claim merely shows that most of their customers were doing an extremely poor job managing bids prior to using their tool. If you have been savvy and diligent, you might not see a performance improvement at all. If you manage other portions of your campaigns poorly, you may see performance get worse.
If a savvy and diligent analyst might not see any performance improvements after letting bidding algorithms take over, I guess we don’t need them, right? Let’s now throw the baby out with the bathwater. The benefits of automated bidding are immense. If I’m managing campaigns spending hundreds of thousands of dollars with over a hundred thousand keywords, I would need to devote hours every day to accomplish the same analysis and adjustments that could be just as easily performed by an automated tool. Now I have less time to focus on all those other factors impacting performance. A good campaign manager really can’t perform his/her job effectively without some kind of bidding automation. Inefficient use of time is invariably the result.
This completes my confession, so let me close with some sensible advice. A bid management tool is like any other tool. It does the job it is designed to do. Expect more, and you’ll be setting yourself up for disappointment. Used properly within a well-structured and well-managed campaign, you’ll reap considerable value. So the next time a bid management software sales rep promises you magical performance improvements, smile and nod. But never fall for such claims. Now you know better.