As an Ad Tech software company providing automated bidding & campaign management solutions, we are frequently responding to advertisers who are concerned that mysterious bidding algorithms are not optimizing their bids appropriately. When I’m tasked with reviewing bid management performance for these advertisers, I frequently find issues unrelated to bid levels that are hurting performance.
Most commonly, I find bidding inefficiencies resulting from campaign structure issues. I think we all intuitively know that structure impacts bidding, but it is something frequently overlooked and forgotten when analyzing bidding performance. The truth of the matter is, campaign structure can render your entire bid management efforts ineffective if setup improperly. So I think it’s time for a refresher discussion of some common structural problems to watch out for.
I’m going to cover two structural issues I seem to run across most often. In this first of two posts, I’m going to discuss how search term matching can inhibit bidding performance. In my follow up post, I’ll delve into the importance of the structural grouping of keywords to assist in bidding on long-tail keywords.
Structural Problem #1: Search Term Matching Impact on Conversion Rate Predictions
The single most common structural problem I find inhibiting bid optimization success is poor search term matching. Note that the most important metric to predict for bid management is keyword conversion rate. We make conversion rate predictions based on an assumption that people who search for ‘big blue widgets’ tomorrow will convert similarly to those who have searched for ‘big blue widgets’ in the past. But we set bids on keywords, not search terms. The two often don’t match, and thus we run into problems.
Consider an education advertiser bidding on the broad match keyword ‘dental hygienist school’. This keyword may be getting matched to searches for both ‘industrial hygiene school’ and ‘dental hygiene school’. These search terms will certainly have vastly different conversion rates. If one keyword is matching to all these search terms, the conversion rate of the keyword becomes the product of which search term Google matches to more often. More matches to ‘industrial hygiene school’ and conversion rates go down. More matches to ‘dental hygiene school’ and conversion rates go up.
In the two charts above, the search term ‘industrial hygiene school’ converts at 0.2% and ‘dental hygiene school’ converts at 9.0%. If targeting a $20 CPL, the first distribution of click volume would lead me to bid around $0.34, and the second would lead me to bid around $1.80. An automated bidding strategy becomes arbitrary because it is unable to exploit that fundamental assumption of historical search term performance predicting future search term performance.
The example of ‘industrial hygiene school’ is one of inappropriate matching, which you’re likely watching out for already. But the more common mismatch is one of misdirected matching. Let’s look at how Google decided to match search terms for a grocery coupon advertiser bidding on general coupon terms (free coupons), more specific product terms (shampoo coupons), and highly specific brand terms (Pantene coupons).
Note that only the third search term for “free coupons” was matched as the advertiser intended. In the first and second search term, Google chose to match to the more general term even though a keyword with a better match was available. Without using more restrictive match types or what I call ‘traffic cop negatives’, you will invariably find search terms being matched to a less desirable keyword. Once again, a bidding strategy will become less effective. Not because we’re matching to a search term we don’t want to show for, but because we’re matching in the wrong place. We’re now setting bids for ‘Pantene coupons’ based on the past performance of ‘free coupon’, losing the value of segmentation.
The obvious solution to these matching problems is a well-planned negative keyword and match type strategy. But work doesn’t end with initial implementation of this strategy. Search query mining must be conducted on an ongoing basis to identify new matching errors. Ultimately, your negative keyword libraries and match type strategies will start to take on a structure of their own. If this exclusionary structure is ignored or mismanaged, you can expect your bid optimization performance to suffer.
Structural Problem #2: Grouping Keywords to Support Long-Tail Predictions
The second most common structural issue impacting bidding performance is a lack of statistically significant volume necessary to make accurate predictions. Yes, I’m talking about long-tail keywords. These tend to be the biggest challenge with bid management, and can inhibit bid optimization in a number of ways.
Let me start with the most severe cases I’ve observed and move to the less severe. I recently audited an advertiser’s account that contained over a million keywords. This account was actually low spending, driving about 30,000 clicks per month. At most, only 3% of keywords had a chance to generate a click, and this would only occur if no keyword received more than one click. They had essentially forced every keyword to be long tail. I don’t care if every keyword was perfectly matched to its exact search term counterpart; this is what I call ‘bid management suicide’.
There is no point even bidding at the keyword level anymore. What do we do when a keyword has only received one click in the last 30 days? Most often, nothing. There’s not enough data to make a decision. But doing nothing with your long-tails is also making a big mistake, which leads me to the less severe long-tail structural challenge.
Take a look at the performance (table 1) for similar blue widget phrase match keywords over the last 60 days. The phrase match keyword “enormous blue widget” hasn’t received nearly enough click volume on its own to predict future performance, but it would be a mistake not to increase its bid. We intuitively know that “enormous blue widget’ will likely perform similarly to all of these other keywords. Conversion rates on the other keywords range from 3.8% to 6.8%. So even if “enormous blue widget” converted only as well as the worst performing similar keyword, we would be willing to pay as high as $0.46 to drive a CPA equal to the average of the ad group. Unless there was a lot more volume with horrendous performance prior to this date range, this keyword is severely underbid.
So how do we formalize what we intuitively know about similar keywords to make better decisions on long-tails? Unless you have access to technology performing some pretty complicated natural language analysis and identifying levels of correlation based on token terms within keywords, you’re likely going to rely on campaign structure. If I were deciding what to bid on a keyword that has never received a single click, I would be smart to start with a bid based on the average conversion rate of the ad group with which it resides. For a keyword with 10 clicks, I might consider both the keyword and the ad group conversion rate, weighting more heavily on the ad group rate. For a keyword with 100 clicks, I might do the same but weighting more heavily on the keyword rate.
Now structure has become a major component of my bidding strategy. The segmentation of ‘yellow’ and ‘small’ widget keywords into separate ad groups now serves a purpose beyond ad association. It helps us make better bidding decisions on long-tail keywords. In short, the structural segmentation and grouping of keywords is critical for effective bid management for long-tail keywords.
Thinking Beyond the Bid
To generalize this topic a bit, no component or metric in your campaigns exists in isolation. Your campaign structure, budget caps, geo-targeting segmentation, number of ads in rotation, ad scheduling, conversion tracking, attribution, and a dozen other variables will affect bid optimization success. If your bid optimization process is only looking at the impact of bid levels, you’re destined for failure, or at least sub-optimal performance.
So the next time one of my counterparts from a bid management software company claims that merely plugging your campaigns into their algorithm will magically improve performance by 30% to 50%, remind them that bid management success isn’t just about bidding. Addressing structural issues like keyword/search term matching and the proper grouping of long-tail keywords for supporting data must be done first. Only then can you get the most from bid optimization, regardless of the bidding tools you are using.