One of the common requests agencies get from large organizations is to provide an estimated return on investment (ROI) for search engine optimization work. Our VP of SEO asked me to explore the possibility of predicting ROI with statistically reasonable confidence. I am no statistician, but I did stay at a nice hotel a few weeks ago, so here is the first half of what I came up with.
I would love for people with a statistical background, or others who have wrestled with this topic, to discuss it at the Search Engine Watch Forums in the thread dedicated to this column: "Just One Agency Point of View."
In order to predict future ROI from SEO, let's look at one way to determine ROI from current efforts. We'll ignore the potential branding value, which is inherently difficult to quantify and would require its own article.
Calculating ROI of Current SEO Campaigns
In the case of predicting ROI for current ongoing SEO, the simplest formula would be as follows:ROI = (Conversions Realized from Organic Listings) – (Cost of SEO)
In this basic formula, the cost of the SEO is a known variable, but unfortunately in many cases the conversions realized from SEO efforts cannot be fully measured. Even with the purest e-commerce site that utilizes the most advanced conversion tracking system, some sales that derived from initial discoveries of the product within an organic search engine results page (SERP) will not be properly attributed to that behavior. This has been described in the past as the "leaky bucket."
Thus, it is easiest to admit from the beginning that one will never achieve perfect ROI tracking of SEO expenses. This actually applies to paid search tracking as well, as there are sales that "slip through the cracks" even with cookie-based monitoring technology that tracks returning visitor behavior over years.
The Power of Web Analytics
The question becomes, "How close to accurate can ROI measurement get when dealing with SEO?" The answer is that it is possible to measure ROI from current campaigns with a statistically high confidence, as long as visitor behavior is properly tracked via a Web analytics system. Currently, only a few such systems can track end-user behavior at a level that is granular enough to be considered accurate. These systems include Omniture's Site Catalyst, WebTrends 8, Visual Sciences (nee WebSideStory) HBX, Google Analytics, and some others.
One of the benefits of working with larger clients is that many of them have access to this type of software. However, we have found that these clients consistently do not have their software properly configured. As a result, we have launched our own service offering to help clients with this problem. That team, as you can imagine, is already forming a waiting list of clients.
Analytics that measure conversions at a keyword phrase level allow marketers to confidently predict future results, given a statistically valid sample. In the past, statisticians have proposed that if a keyword phrase yields 50,000 impressions and 500 click-throughs, the ensuing prediction can be considered sufficiently accurate. Thus, in order to adequately predict future performance, at least this amount of data should be collected.
This poses a problem for longer keyword phrases, also known as "long tail" terms. In many cases, these types of keywords will yield minimal activity in comparison, averaging between 100 and 1,000 searches per month. Although the conversions generated from these phrases can be included in the aggregate SEO ROI, it is very difficult to confidently predict future conversions at a per-keyword-phrase level. This will be discussed next time.
How Research Conducted in 2007 Will Help Improve ROI Attribution
We are currently able to report conversions yielded from organic searches, given the availability of data from one of the previously mentioned Web analytics tools. This allows the SEO team to determine which keyword phrases perform best. When available, this data can be compared to paid search statistics for the exact same keyword phrases, and the two should often match up fairly closely.
Research analysts at JupiterResearch, as well as with MarketingSherpa, Hitwise, and other competitive intelligence providers, hope to be able to better understand the relationship between paid and organic searches. Plans for 2007 include identifying searcher behavior when faced with multiple listings for the same site/domain.
Does having both paid and organic listings (or even a greater variety in the new Google Universal Search such as those generated by local, YouTube, image, shopping engine feeds or Co-op results) help increase conversions on an aggregate level? Does having multiple listings somehow affect the ROI derived from either the main paid or organic campaign? These questions will hopefully be answered through research conducted in the coming months. Analysis of the findings will also help to more confidently predict future ROI for current campaigns, and for those not yet launched.
Next time, we will discuss the major problems of trying to estimate future ROI derived from current or proposed SEO efforts. These issues are very important for organizations of all sizes to comprehend when evaluating SEO proposals.