Measuring ROI for current campaigns is a piece of cake, right? But how about predicting the ROI for future SEO campaigns? Navigate the waters with Chris Boggs as he discusses the challenges of dealing with the unknowns -- not easy, but doable.
Last time, we said measuring ROI can be difficult for SEO. Again, I am no statistician, but based on the challenge of predicting the ROI for future SEO campaigns, you may feel like measuring ROI for current campaigns is a piece of cake.
Can we confidently predict the ROI for future SEO campaigns? Well, maybe. The simplest possible formula that could be used to predict ROI for proposed SEO campaigns may look like this:
Predicted ROI = (Anticipated revenue from SEO efforts) – (Proposed cost of SEO project)
Unfortunately, it is not that easy to come up with these variables. Tackling “anticipated revenue” first, we encounter a variety of problems. Revenue should take into account all conversions from traffic derived from organic listings, as when we measure current SEO.
One problem is predicting the number of searches for each organic listing to be optimized. The next major issue is predicting the actual ranking – number one, two or five – and how to match the potential click-through rate (CTR) based on unknown positions. It is highly likely that a number one ranking returns a greater CTR than number five, but the number five must nevertheless not be ignored. Additionally, how can you assign a specific value to some types of conversions? Lastly, the effects of SEO may not be fully realized until well into the second year or beyond. How are these “latent” conversions accounted for in the prediction?
Although there are many keyword prediction tools, the primary tool used by our team is WordTracker, which provides an estimate of daily searches for specified terms. We know that this is not the most dependable prediction, but it can act as a guideline for many general terms. Some account managers choose to ignore these numbers, especially when analyzing SERPs for terms and finding multiple paid search competitors that are bidding exact or standard match instead of broad match. If there are people bidding on a long tail term, chances are that it drives traffic, regardless of the fact that WordTracker may claim zero searches per day in its predictive function.
There are other more robust keyword estimators, a category known as “competitive intelligence tools,” such as Keyword Max, Trellian, Hitwise, and qSearch from ComScore. It would likely be a wise investment to buy a license for one or more of these products should you attempt to predict ROI for a proposed SEO campaign. The more predictions you use to develop an average, the more accurate your number of predicted searches will be.
Using paid search reports has proven to be one of the most reliable forms of determining high-value terms to target for SEO. Unfortunately, not all clients will have the luxury of running full-exposure budgeted campaigns for a significant length of time to gather the amount of data needed, especially for waves of searches for products or services that ebb and flow with the seasons.
Another thing to consider is that after the discovery phase, some of the unidentified long tail terms may end up getting ranked, thus driving conversions. These rankings should rightly be considered a result of the SEO campaign, even though all terms were not identified in advance. Thus, padding should be added to the potential aggregate ROI in order to improve accuracy.
The next problem in predicting anticipated revenue is that it is hard to assume a CTR. As discussed last time, the possibility of additional listings for the same site within paid or local results, affects its actual position in the top ten, etc. skewing the click-through rate dramatically. Without established rankings and a track record of click-through percentages, there may be a large gap between predicted CTR and actual CTR. Simply taking an average CTR for the number one through ten results on the first page of the SERP may be the only solution for this part of the equation.
It may be wise to consider researching aggregate CTRs for clients that are maintaining rankings in the top ten. This would be possible if the paid search impression data for the particular term were available and compared to the referrals from organic listings for each of the same keywords. If we did this for a statistically valid sample, we could probably come close to predicting CTR based on actual position within the top ten. Of course, we can probably pay for this kind of data as well.
Some conversions are more valuable than others. In the case of an eCommerce site, should a conversion that yields a $100 sale be considered more valuable than a conversion yielding a $10 sale? The obvious answer to this question is “yes.” This brings us back to the choice of predicting ROI on a per-keyword-phrase level versus an aggregate prediction, which would force assumptions of how many $100 conversions were achieved versus $10 conversions. We could base this on an average sale, but may end up short-selling ourselves.
If a site’s conversion event is a signup for a newsletter or an insurance quote, the site owner needs to assign a value to each visit. Again, this is difficult, as some searches are more likely to yield conversions than others. Using an insurance lead as the example, the same search for “health insurance Philadelphia” could yield a family of four buying the best available individual plan, or a single person buying a low-cost health plan.
Once again, an arbitrary estimate would be required for the average sale. The insurance example is even harder to predict due to the fact that some clients signing up may stay with the broker for years, continuously increasing the ROI from that one lead. Others may drop the plan and its commissions after just a few months. There are probably many other similar scenarios that could skew predictions based on a highly variant lifetime value of the client.
In the case of sites that do not have an online conversion mechanism, we can’t overlook the branding factor in discussions of ROI derived from SEO. This is particularly important and would require a complete analysis on its own merit.
On a positive note, we are able to identify and research Web site competitors, enabling us to estimate the degree of difficulty in attaining top ten rankings. The problem is that it takes at least three or four hours of solid research to accurately predict how difficult it may be to eclipse the sites in the top ten. It can be done, but would this be justified as a discovery cost?
The idea of an “assist” exists, where someone may search a more general term and find it organically, early in the buying cycle, and then end up searching a branded term and finding a paid listing, converting on that. This is similar to the leaky bucket theory discussed last time. Should half or more of the credit be given to the organic result that the visitor first found, or even other results found within subsequent searches — all of which arguably reinforced the brand?
This leads nicely to the question of “latent conversions.” What happens if we SEO a site and the client ends the relationship after a year, subsequently gaining considerable rankings in months 13-24 and beyond, and the site is driving previously unrealized conversions as a result of the SEO work? This is why many top SEO consultants are moving aggressively toward a pay-per-performance clause within contracts.
Incentive clauses would really be the easiest solution to the problem of predicting ROI, especially if the deal hangs on the client wanting such an estimate. If the client and agency are willing, the frontloading of costs could be minimized to reflect actual setup work completed, and ongoing work could be paid for – after a certain time frame – by conversion-driven incentives. The most likely scenario for moving toward this model would require at least a two-year commitment and verbiage in the contract calling for a buy-out.
To recap, the major problems with predicting ROI for SEO include the uncertainty of most data sets and the inability to determine the actual ROI over the life of the Web site. Following is a list variables that should be considered in order to predict ROI.
We started with the following formula:
ROI = (Anticipated revenue from SEO efforts) – (Proposed cost of SEO project)
Breaking this down further, we have:
(Anticipated revenue from SEO efforts) = Conversions derived from Organic search visits
The unknown and known variables that can help solve the formula are:
Unknown:
Known:
Obviously, there is a lot of research that goes into predicting ROI of future SEO campaigns. Would this type of cost ever be justified prior to actual contracted work?
At least:
(Proposed cost of SEO project) = Known (unless performance incentives are used)
Paid search data, when available, would be very beneficial in properly estimating future ROI from SEO efforts. However, based on the required information needed to accurately predict conversions/value driven from an unknown number of keyword phrases over an infinite time period (as long as the site exists), I do not believe that it would be worth the effort to try this. The best solution, in my opinion, would be to structure performance incentives into contracts where clients insist on having an estimated ROI.
I look forward to learning more as people continue to discuss this analysis at the SEW Forums in the thread: “Just One Agency Point of View.”