Controlling Google's Inconsistent PLA Target Serving

Many moons ago, RKG's George Michie wrote a post lamenting the fact that Google will often serve broad match keywords with higher bids over an exact match term that matches the query but has a lower bid. This phenomenon shouldn't happen according to Google's own literature, which details that a keyword should only be served in the place of a more closely matching keyword if it has a lower bid and higher ad rank. Our own experience has proven this to be inaccurate, however, and diving into the way Google's Product Listing Ads autotargets are served, we find much the same story. Different levels of PLA autotargets act in much the same way as matchtypes. A Product ID target can only be served if the product shown through PLAs exactly matches that ID, much like an exact match keyword can only be served if the query exactly matches the keyword. Broader targets, such as All Products, adwords_grouping and product_type behave more like broad match keywords, with the granularity of the target controlling the scope of how many products can be served through these targets. As such, it would seem that the optimal strategy for PLAs would be to have as much traffic going to Product ID targets as possible, so that bids are calculated at this most granular level. However, this isn't as simple as it sounds. Higher Bids Rule Most of the Time Google recommends that advertisers keep an All Products target active in their PLA campaigns as a catch all to 'ensure that all of the products in the product feed are eligible to show' through PLAs. At RKG, we've found that if Product ID bids are carefully monitored and kept above the bids for broader targets, such as All Products, that the product may fall under, the Product ID targets can get close to 100% of the traffic that goes to the products that have these targets. This isn't always the case, however, as there are instances where Google seems to ignore Product ID targets regardless of bids. For example, a Product ID target for one advertiser was launched with a bid that was 85% higher than the All Products target, yet Google ignored the more granular target and instead sent hundreds of clicks a week to the All Products target. This is problematic for a couple of reasons: 1)      The reason the Product ID was launched with a higher bid was because this was a well performing SKU and we wanted more PLA traffic on it. Funneling this traffic to the All Products target with a lower bid keeps us from being able to dial up traffic on this particular item. 2)      The All Products target was absorbing all of the well performing traffic on that Product ID, which performed at a significantly higher sales per click than the All Products target as a whole. Thus, the apparent performance of the All Products target is inflated. While instances like these appear to be the exception, it still doesn't make very much sense that the broader target would be served with a lower bid, and draws into question how well Google is able to assign quality score to a Product ID target. This is especially true of accounts that have long standing active All Products targets when the Product IDs are added. As Google has more history for the All Products target, this target is assigned a higher quality score and garners traffic that could have gone to more granular targets. In order to combat this, RKG has found that by slowly raising the bids for more granular targets while at the same time reducing the bid for the All Products target, traffic will slowly move over to the more granular targets. No Point to Granular Targets with Low Bids? Most paid search advertisers aim to hit a certain cost to sales ratio as their primary efficiency metric.  If sales per click goes up, the advertiser can afford to bid more at the same efficiency, and they will drive more traffic. This carries over to PLAs as well. Similar to keywords, though, Google will often choose to serve less granular targets if the bids are higher than those for more granular targets that could be served. In situations where a particular product is getting a lot of traffic under a broader target but not performing as well as other products in that target, it would be nice to be able to launch a Product ID target with a lower bid to maintain efficiency for this traffic. With the target cascade method of serving that Google is currently using, however, this seems impossible as the broader target with the higher bid will absorb the traffic while the Product ID target is ignored. Lack of Control for Shifts in Demand Shifts in consumer demand for a product can drive the PLA performance of a particular product up or down quickly. Even if an advertiser launches a Product ID target with a higher bid than those for more general targets because the product is performing well, performance for that product could become worse than that of broader targets at any time. In this case it would be best if the Product ID target could be bid down below the broader targets in order to maintain the return on ad spend for that product. This traffic would mostly just go to the broader targets, however, rendering the Product ID target useless unless performance picks back up and allows the advertiser to increase the Product ID bid back above the broader targets. Controlling Traffic with Negatives In the case of broad match keywords superseding exact match terms with lower bids, one solution that works fairly well is to apply exact match negatives for all exact match keywords to broad match terms. This eliminates the broad match terms from being triggered for those searches. Negatives can be used similarly with PLA targets in order to prevent different targets from showing on particular queries for which an advertiser would prefer to show a different product. For example, if an advertiser were selling the new iPhone 5c and there was a Product ID target, it would make sense to add negatives such as 'iPhone 4' to the ID target to prevent that product from showing to people searching for the older model. One aspect where the keyword matchtype to PLA target type analogy differs, however, is that a Product ID can be triggered by multiple relevant queries, whereas an exact match term can only be triggered by one query. This makes it more difficult to control traffic for a product from going to a broader target as many different negatives would need to be added. These negatives would also have to be added not only to the All Products target, but any category level targets that the Product ID may fall under as well. In the iPhone 5c example, this product could be triggered by queries as broad as 'cellphone' or 'smart phone.' In order to keep an All Products target from showing the iPhone 5c product for which there is an ID target, negatives for these broader queries might have to be added to keep that product from showing through the broader target. Further, if there is a category target such as 'iPhones' that the iPhone 5c also falls under, those negatives would have to be added for this target as well. Thus, while negatives do help to control which targets get traffic from different keywords, they are difficult to implement if the goal is to shift traffic for a product from a broader target to a more targeted one. Conclusion If you're going to launch more granular targets for PLAs, whether they be more specific category targets or Product IDs, it makes the most sense to launch only those that you have the ability to bid more for than the more general targets that the products in these targets could fall under, such as All Products. Otherwise, the traffic will, for the most part, just be headed to the broader targets. It would be nice if Google provided some way to prevent broader targets from getting traffic on products that have Product ID targets launched. This could either be through back end adjustments that Google makes to ensure the more granular target is respected, or by allowing advertisers to block specific IDs from being shown through broader targets with ID negatives. As Bing begins to flesh out their own product ads, I hope they'll consider building such functionality into their approach.
Join the Discussion