Data Mining: The Heart of Analysis

Over the last few years, so much of the talk in our industry has been about tracking ROI, and how it’s imperative to be able to do so before spending a single dollar on paid search. Agreed.

So with all that talk, it would seem obvious that by now everyone can not only can track their ROI, but that they also know exactly what their returns are. Unfortunately, that’s only happening in our dreams, as a study we published just 18 month ago showed that 31 percent of search marketers do not, or cannot, measure the ROI of their search marketing efforts.

The reality is that many marketers are good at squeezing every penny out of a paid search program. However, what do you do when the needle won’t move anymore — when you can’t invest another dollar without diminishing your returns? The answer lies in the data. There is so much analysis that can be done — and everyone claims that they do it — but I wonder if that’s entirely true.

Would you consider identifying your top-performing keywords as “analysis?” Maybe. Would you consider identifying the time of day or the day of the week that’s most opportunistic to your performance as “analysis?” Perhaps. While these factors are hardly advanced, they are definitely required; but if you are looking to push the needle, you’ll need to dig deeper. Here are a few approaches that offer insight beyond the basics that are relatively easy to implement, and yet are all too often overlooked.

Start With the Basics

Begin by thinking about the user behavior patterns that exist between queries and sites, and about the correlations between various online initiatives (not just paid search) and offline marketing activities. Which initiatives are driving which programs’ results? Which programs’ success depends on other programs? Are your TV and radio ads effectively driving prospects to search for you online? Are your search programs leading consumers to call your call center or walk into your stores?

There are also some fundamental and commonly understood steps that can be taken to gain efficiencies and improve paid search advertising performance:

  • Improve your call to action. Improving your click through rate (CTR) through a more persuasive call to action in your ad will allow you to gain higher positions within the sponsored listings, which may yield yet more volume.
  • Settle for second place. Decreasing your bid to lower your position in the sponsored listings can save click costs without significantly diminished returns.
  • Focus on the funnel. Improving your landing pages and conversion funnel can help to gain a better response rate and increase returns.
  • Look to the tail. Increasing the number of long-tail keywords on which you bid can typically produce traffic from less expensive clicks.
  • Go beyond Google. Expanding your bidding to tier-two engines or content networks can increase your reach to a broader audience.

Just by getting the basics in order, you should see some improvements. But when you take the next step — to actual analysis — there are creative and complex patterns that you may be able to observe that will help you better understand ROI and hence drive greater performance.

Then On to the Analysis

Before you begin any type of analysis, make a list of questions you’d like answered, or perhaps suggest a potential theory that you’d like to prove or disprove. You want to look for those answers that would make you to do things differently. This approach will help guide your analysis. However, it’s critical that paid search NOT be examined in a vacuum. Too many marketers treat the channel solely as a direct response mechanism. Granted, this might be necessary from a financial perspective, but there is other valuable information in the data if it is mined and analyzed properly.


There is typically some form of relationship between your audience’s keywords. Many in the industry claim that searchers first search on broad phrases then narrow their query to become more specific; hence marketers shouldn’t discount broad terms just because they don’t look favorable from a ROI perspective. Clearly, there is some truth to this claim, but so too is the opposite — specific queries can lead to broad searches. But don’t take my word for it, look at your data!

By analyzing the cookies of your site visitors, exclude from your data all those who have only clicked to your site from an ad once and then have a look at the statistics and trends of the queries launched and ads clicked by those visitors who performed multiple queries. This will not only help you understand what keywords to buy, but also affect how you attribute cost and returns between keywords.

You should also look at the ROI value of early-stage keywords, by attributing the eventual conversion to the first keyword a user queried and clicked on to arrive at you site. Then compare that to the ROI value of keywords if you were to attribute the eventual conversion to the last keyword queried immediately prior to converting. This type of analysis goes by many names such as: Multi-Query Analysis, Search Funnel Analysis, or First versus Last Click Analysis.

There is also insight to be found by analyzing the search properties that your audience may be visiting. If you’re running campaigns in multiple engines and ad platforms, you should examine the behavioral patterns of your audience segmented by source. You may find that there is a greater propensity for a user to click to your site and convert if they saw you on search engine XYZ.

For example, imagine for a moment that you observe that 10 percent of your conversions that came from Google had also visited and clicked on an ad in Yahoo. What would happen if you throttled Yahoo back? Surely, that would be nice to know before you negatively impact your overall ROI by neglecting to take Yahoo’s contribution into account.

It would also be a good idea to observe the effects on conversion when ad messaging is kept consistent or varied across paid search sites. Again, too many people focus on ROI per engine, per keyword, without looking at any of the correlations between the elements.

Making it Happen

Obviously, tracking is a key factor in being able to conduct this type of analysis, never mind having enough in-house resources and the skill sets required to perform the analysis. And as good as you may be at managing paid search campaigns, it would behoove you to have a statistician or data miner analyze the data to find the interactions of these touchpoints and report back to you, especially if the data set is large. Ultimately, what you should find is a set of scenarios and circumstances that will guide you to either prevent losses and/or capitalize on newly identified opportunities.

Remember, there is analysis, and then there is analysis. What you reap from it depends on what you put into it. If you are looking to push the needle, you’ll need to go beyond the basics. The tips discussed here barely scratch the surface when compared to what is possible. The analysis becomes even more interesting when you examine the correlations between online and offline channels, especially when you consider that online should mean a lot more than just search. But that will have to wait for a future article. Until then, happy data mining, and yes, in case you’re wondering, I am a data geek at heart.

Related reading

A screenshot of visual search on Pinterest. On the left is a picture of a copper angle-poise lamp, with the words 'Visually similar results' above it. Down the right-hand side are a number of pins showing similar lamps.
google home
Simple Share Buttons