An advanced paid search bidding algorithm will consider a number of attributes in effectively estimating the value of a keyword. In past research, RKG has found that item price can be a strong predictor of keyword performance and that price data should be included in the signals assessed by your bidding system. Why?
Simply put, sales per click (SPC) varies by item price, although not necessarily exactly how you might expect. To demonstrate the variation in performance by item price, the graphs below break out product level terms for four different PPC programs into price buckets according to the price of the item. The price buckets are plotted against the observed SPC:
Note: We consider a keyword to be a product level term here when that term’s landing page features a single product and when the term itself describes only that product. For example, the term “Sony television” would not be a product level term because it would apply to a group of products, whereas the term “Sony KDL40BX420” refers to a specific model and would likely be taken to a landing page for that model.
First, for all four advertisers above, SPC clearly varies significantly across price buckets. As such, these advertisers should be willing to pay far more for clicks that lead to product pages in the price buckets with the highest returns per click.
Another trend that holds true for all four advertisers is that the average order value of terms in the highest price bucket was lower than the lowest price in that bucket. This means that for these advertisers, terms for their most expensive products are frequently not bringing in orders for those products. This speaks to the potential of incorporating links to more economical alternatives on landing pages for expensive items.
The decline in SPC around high-ticket items may also be amplified by a greater likelihood for customers to order through the call center or purchase in-store. While RKG offers a call center tracking option, that data was not considered for this analysis.
While the seemingly normal distribution of these charts may make estimating price bucket performance seem easy, some trends vary from advertiser to advertiser. For example, the highest SPC price bucket for three out of the four advertisers contains their average order values for product level keywords, while the fourth was an exception.
For two of the four advertisers here, SPC veers from a normal distribution to go up in the highest price buckets. It’s important that your system have the ability to recognize these sorts of anomalies on the fly and be able to account for them in calculating bids.
Product Price vs Category
As with any other bidding signal used to set keyword bids, your bidding system should be able to determine the importance of product price for your specific program. However, for some advertisers, it may be the case that something like product category or subcategory could be a better predictor of performance.
For example, if being part of a certain product category, say t-shirts, is a good indicator of an individual product’s price (in that most t-shirts sell for roughly the same amount) bidding by a category signal may be more effective than bidding by a price signal alone. The category signal will take into account not only the price of the item, but other factors that affect demand for that category such as seasonality.
On the other hand, a category such as Sony televisions may include products with a wide range of prices and varying levels of demand. In this case, it seems that the price could be a better predictor of performance than the category. Which do we choose and how can an advertiser make these determinations at scale?
Bidding The Right Way
While it’s still not uncommon for advertisers to establish bids at the adgroup level without taking other signals into consideration, this methodology is not ideal. A sophisticated bidding system will have the ability to measure the importance of multiple signals at once and incorporate them into the bids of each keyword regardless of account structure. Bidding signals lie in every aspect of every product sold and are very rarely accurately summed up by the adgroup to which the keywords belong, which is often just a product of website hierarchy.
It is vital that your bidding system not only be able to take into account a wide variety of signals, but that it also be able to determine which of those signals are the greatest predictors for keyword performance.