AdWords and Bing are great at finding in-market consumers (consumers currently researching or looking to buy a certain product), but they don’t provide you with many tools or capabilities to effectively segment or target different types of consumers like you can do in display advertising or on Facebook.
However, both AdWords and Bing allow you to target and bid at the zip code level using geo bid modifiers. If you think of zip codes as a way to represent audience, you can leverage AdWords/Bing geo bid modifiers capabilities to target and optimize for different audience segments in paid search, creating thousands of new ways to target and optimize performance of paid search campaigns!
The following article answers some frequently asked questions and guides you through how to use geo bid modifiers to hyper target your campaigns and boost paid search performance.
AdWords and Bing allow you to set a higher or lower bid by zip code, at the campaign level. For example, if you have a maximum bid of $1 on a keyword, and you set a +25% bid modifier for zip code 90210, then for that zip code your maximum bid for the keyword will be $1.25.
There are approximately 42,000 zip codes in the US, ranging from hundreds of people to tens of thousands of people per zip code. Think about the zip code you live in – you and your neighbors may have many things in common such as household income, age, presence of children, ethnicity, and even the kinds of sports you are interested in or the programs you watch on TV. And these characteristics can be very different from people in another zip code 1000 miles or 10 miles away from you.
Unfortunately it’s not that easy. Recall there are approximately 42,000 zip codes in the U.S. To have enough statistically significant data for every zip code, you would need hundreds of thousands conversions per month, for each of your products! The reality is, no retailer has this kind of scale by product, not even Amazon!
For example, below is the conversion distribution by zip code for an online furniture retailer with approximately $1m of paid search spend over a 6 months period. The “Sales” metric indicates the number of conversions. As can be seen on the map, the vast majority of zip codes (in a densely populated area like the LA metro area) have 0 conversions, with a handful of zip codes accumulating one or more conversions during the period. The zip codes with 0 conversions are not all underperforming zip codes – most of them just don’t have enough click volume to generate any conversions. Therefore, zip code performance alone is not a reliable method to predict future performance due to data scarcity.
The secret is to use external data points to help you find similarities between zip codes. For example, suppose you knew the percentage of people with a Bachelor’s Degree by zip code, you could aggregate all zip codes based on Bachelor’s Degree percentage, and then look at the conversion rate or return on ad spend (ROAS) on an aggregated basis.
The following graph shows the correlation between percentage of people with Bachelor’s Degree and ROAS and the correlation between Adult Smokers and ROAS, calculated by zip code. As we can easily see, the ROAS increases if there are more people with a Bachelor’s Degree in the zip code and decreases the more Smokers there are in the zip code. This implies that you can raise the geo bid modifier in zip codes with lots of people with Bachelor’s Degrees and less Smokers, and decrease the geo bid modifier in zip codes with less people with Bachelor’s Degrees and lots of Smokers.
The previous section described a very simple example of geo bid modifiers optimization using two dimensions (bachelor’s degree and smoking). But these data points weren’t picked randomly, they were chosen because they showed a good correlation with conversion rates and return on ad spend.
In reality, there could potentially be thousands of data points that can be used as dimensions for optimization. Most of them however, would not show any correlation to your main KPI, a minority will show a small correlation and only a couple dozen or so will show strong correlations like we saw in the example above.
So how does one find the “right” data points to use that will show strong correlation?
One thing you can try doing is to using your prior knowledge about your target market or just common sense to select data points. For example, if you know that your product is geared towards more affluent consumers, perhaps household income might show strong correlation and could be a good optimization dimension to use. Once you have a thesis on what data points might be relevant, the next step is to find and acquire the data point from one of the data providers, cleanse and normalize the data to your needs and then evaluate the data points relative to your KPI using a statistical regression model.
At cClearly, we’ve already done all the hard work. We aggregated, cleansed and normalized thousands of zip code level data points including demographic, financial, behavioral, attitudinal, psychographic and firmographic datasets.
We take your performance data at a zip code level (available through an AdWords/Bing report or through your back end system if you use another system to track conversions), fuse it with thousands of zip code level data points from our unique datasets and automatically test each data point against your main KPI (such as sales or return on ad spend) to identify the best correlations.
As a final step, our system uses those thousands of zip code level data points and the many correlations it finds to predict a conversion rate and return on ad spend by zip code, based on which it calculates a recommended bid modifier by zip code.
For example, below is the same map as before showing the number of sales by zip code for a retailer in the LA metro area and the data scarcity challenge. Immediately below it, is the map showing the bid modifiers our system is recommending for the same retailer. As the map shows, while many zip codes have no conversion data in the first map, all zip codes include a recommended bid modifier in the second map, created through the use of thousands of external data points.
By bidding up zip codes that are predicted to perform better and bidding down zip codes predicted to underperform, you are essentially shifting your ad budget away from under-performing users and shifting it towards your best performing users. In other words, your ad will show up more frequently and in a better average position to the people most likely to buy from you.
Shifting budget towards your best potential customers results in improved conversion rates, reduced cost per acquisition, an increase in return on ad spend and an increase in overall performance for the account. Overall results can vary significantly, but it is not atypical to experience a 20%-30% increase in paid search performance when implementing an audience-driven optimization approach.