SEO is both an art and a science, as with any scientific discipline it requires rigorous testing of hypotheses. The results need to be reproducible and you have to take an experimental approach so as not to modify too many variables at once. Otherwise you will not be able to see which change is responsible for the result which would create a problem. 

Although you are able to gain a tremendous amount of knowledge of the best SEO practices , latest trends and tactics from SEO blogs, forums, and e-books it is hard to separate the wheat from the chaff and to know with any degree of certainty that an SEO related claim will hold true. This is the point when the SEO testing comes in, proving what works and what doesn’t.

Unlike multivariate testing for optimising conversion rates, where many experiments can be run in parallel, testing requires a much more serial approach for an seo company. Everything has to be filtered through the search engines before the impact can be gauged.

This has been made more difficult by the fact that there’s a lag between changes and the revised pages getting spidered, as well as another lag while the spidered content makes it into the index and onto the search engine results pages or SERPs.

On top of all of this the results delivered depend on the users search history, the Google data centre accessed and other variables that may occur that you cannot hold constant.