Confidently Projecting ROI for SEO: Can it be Done?

This spring marked the annual Search Engine Strategies Conference in New York. I was fortunate to be invited to present a session on the topic of measuring ROI for search engine optimization (SEO) efforts.

In 2007, I wrote for SEW that it was difficult to fully measure past SEO impact as well as to confidently predict future ROI for SEO. During the SES session, I discussed the many variables had changed since 2007, and three specific ways to model SEO growth in a relatively confident manner.

I did spend a fair amount of time covering the obstacles to accuracy, as noted within some of the feedback I received from the audience that “I made it seem impossible.” That was not my intention.

When discussing statistical confidence, I fear I’m not well trained, but I do know the difference between highly predictable sets and those with more variables and, thus, lower confidence.

My point is that to confidently measure ROI for SEO, you have to temper expectations of high accuracy within exact number of visits or conversions, and lean more toward the school of performance growth as measured by trending lines. The models which I will discuss below have worked well and have proven to be in some cases surprisingly close to numerically accurate, and have certainly shown the ability to be trend-predictive.


Some argue that it’s easier to predict weather than to project organic search performance. However, for some it can be a breeze! Why is this?

As with most things search-related (and social and display media for that matter), the answer lies within the industry and specific niche. Some keywords have very predictable volumes, and a fairly consistent competitor set which can lead to “average” click-through-rate (CTR) that is consistent and subject to seasonality and other niche trends.

The arrival of universal search and the further “mucking up” of the old school 10 links search result page with local, products, realtime, etc., has caused other popular keywords to lead to volatile exposure and limited real estate.

CTR is the most difficult thing to accurately predict, thus a large sample of keywords that a site has potential to rank for is crucial to more confidently predicting growth from natural search.

SEO ROI Model 1

In 2009 I worked with our analytics team to come up with a projection model based on previous traffic to a domain. The ecommerce site already experienced thousands of visits, but we felt the new platform and additional optimization would yield positive growth.

We took the average monthly traffic and then based projection on growth from that average.

The problem with this model was that it used average traffic and did not take seasonality into consideration. However, it turned out to be far closer to accurate especially after the initial six months post-launch.

This is very rudimentary but works if you only have a year’s traffic.

SEO ROI Model 2

The next version for another large ecommerce client that was undergoing a platform change used actually monthly traffic averaged over a two year period as the baseline to measure against. These numbers were then run through a multiplier, based on different levels of SEO to be performed, which yielded the predicted traffic for the specific months of the upcoming year (or longer).

The variables and limiters in the multiplier are based on two things:

  1. Agency/personal team member experience working on sites within the industry (the “easy” as well as obvious climb to a certain level based on how low the baseline is, and past experience with the platform or for keywords with similar level of competitors).
  2. Total predicted search volume multiplied by potential CTR for conservative to moderate page exposure (this one involves its own multipliers for long tail, or using less trustworthy “broad match” keyword volume estimates).

The second model proved to be highly accurate, given uncertainty over level of implementation that would be achieved. One SES attendee was unhappy I couldn’t give all the multiplier variables and limiters, but frankly the above doesn’t take much more to start inserting numbers into before you reach your comfort zone.

SEO ROI Model 3

This one was developed by Marios Alexandrou, our director in the New York office. This method involves using competitor traffic estimates to show potential, especially for non-branded keywords that are very difficult for pharmaceutical sites to rank for without properly leveraging SEO.

The key is finding the competitors that are doing it right. Pharmaceutical companies love to see precedence in order to feel that a goal is attainable, so this model provides that as an extra benefit.

The final output shows a potential volume of traffic that one or more competitors are experiencing from a small sample of popular and relevant keywords, based on using tool estimates and a standard CTR.

Next Step?

The next step in predictive analysis for SEO performance involves self-correctible projections. As marketers increasingly consume real time data, provided by exponentially more powerful analytics platforms, one day SEO could be at a point that PPC, email, display and other online marketing tactics have attained: highly confident ROI projections.

For where we are now, the models described provide the closest to confident projections for SEO available in 2011.

Related reading

serps of tomorrow
2018 SERP changes impact SEO