PPC and SEO Integration: Modelling of Search Efficiency
Data & Analytics

PPC and SEO Integration: Modelling of Search Efficiency

Milan Nayee • 05/09/2018

Background
Meaningful and tangible channel integration has long been the goal of many a digital marketer. Nowhere is this goal more commonly sought after than in search, such is the obvious overlap of the paid and natural disciplines as well as the combined experience for the user within the search results page (SERP). There are a variety of routes to integration between these channels: the testing of messaging and title tags, the content and structure of landing pages, and conjoined keyword research and targeting strategies. But the purpose of this article is to focus in on one area in particular; the balancing of each channel’s respective rank to drive overall efficiencies between the two. Fundamentally, in highly competitive verticals the costs of paid search can be significant, so is there a route to alleviating this by modelling combined SEO and PPC performance? Can we find scenarios where we can dial down paid investment that has a minimal impact on overall search performance?
The dynamic between the two is complex. While there is consistent fluctuation of rank within each, and the format of the SERPs can be inconsistent for different search categories, the premise of our analysis was that we could more easily control the positioning and exposure of paid search – moving it around the (supposedly) less volatile SEO results.
 
Case Studies
A study by google revealed that paid search ads accompanied by an organic search result on average only occur 19% of the time. And only 9% of the time a paid search ad shows with an organic result in position 1.
 

 
                                                                                                                                       
The boxplots above are constructed by taking the Upper and Lower Quartile of the y-values for a given Organic rank which creates the box. The median is the point inside box, and the end of the whiskers are either the max/min values, or in a position such that any point further than this are outliers.
Figure 1:

  • When an organic result didn’t appear there were on average 66% of ad clicks
  • As the organic position decreases, the percentage of ad clicks decreases too.
  • When combined with a boxplot of percentage of ad impressions, which exhibits a similar behaviour as the ad clicks, we see that the CTR of ads are higher when accompanied by an organic result, and the ad CTR decreases as the organic position lowers.

Figure 2:

  • Shows the distribution of ad positions with respect to its organic rank
  • The PPC position is lowest when there is no organic result, but highest when the organic ranking is highest. However the ad position decreases as the organic rank lowers.
  • This does not imply they influence one another, this is simply showing the behaviour of the advertisers.

Given that ad CTRs increase with ad position, which we can control to some extent with our max CPC bid, and the occurrence of higher ad positions when there is an organic result (and possibly an even higher ad position when there is an organic result in rank 1) is a likely explanation for the CTR behaviour – we investigated this area further in our study.
 

  Figure 3

 
Not surprisingly, the results above show that the lower your organic ranking, the greater the percentage of incremental clicks from PPC. However, what is surprising to many of us is the growth in incremental clicks when they appear together. Even the first organic ranking can benefit from an accompanying ad. There are vast amounts of research based on incremental clicks and its impact on the cost of ads, such as pricing models that focus on cost per incremental action (CPIA). We will focus on when a Paid Search ad is present and how its SEO ranking affects it’s overall CTR and cost.
 
Methodology
We used data from one client, and it was important that they are in a highly competitive vertical because this gives us a good spread of data. To utilise both SEO and PPC data, from the data available, we pulled it separately from two different sources, namely Google Search Console, and Google Search Ads. The data was collected over seven and a half months, and segmented daily. As an analyst we spend 90% of our time cleaning data, and this is where the journey begins. Firstly, we had to ensure both data sets were compatible to allow us to fuse the data together. With multiple keywords and phrases presented in the data sets, it is reasonable to filter for exact match only, because the most precise data lies here, but more importantly ensuring the data sets can be merged together without a loss of accuracy. With different measurements presented from either platform, the KPI’s we focused on to study were:

  • Position (PPC & SEO)
  • Overall CTR (PPC + SEO)
  • CPC (PPC)

This can be seen in the table below:
 

Figure 4

 
By combining the data sets together we can see clearly that for any given day and keyword the SEO & PPC position, their CTR (separate and combined) and the CPC. Now that the data is gathered together we can formulate some hypotheses:

  • Hypothesis 1: There exists some point X such that the incremental increase in PPC cost has little/no positive improvement on the combined CTR.
  • Hypothesis 2: There exists some identifiable SEO & PPC position other than 1:1, where it is more effective to be, in terms of combined CTR.

These hypotheses both have the common goal of saving money whilst keeping performance optimal, to reinvest the money saved elsewhere in the marketing business.
Our aim to prove Hypothesis 1 is to create a graph that maps the PPC position against the CPC, for a given SEO position. We would like to see the graph shape to be convex, and turning at some point X, after which the graph plateau’s. This is known as the point of diminishing returns (PoDR), after which further PPC investment would show a negligible increase in overall CTR.
To prove Hypothesis 2, we will create a heatmap to show which areas are efficient/inefficient, this will be the best way to highlight the important areas to focus on.
 
Results – Hypothesis 1

  Figure 5

 
For all keywords we plotted the average Paid Search CPC each day, split by differing SEO positions (1.0 – 1.9, 2.0 – 2.9, 3.0 – 3.9 and 4.0+) and the combined CTR. These regression lines were created in a such a way to reduce the most error between datapoints. You can see from the graph that there exists a turning point, known as the PoDR.

  • For SEO position 1.0 – 1.9 (red line) it is worth increasing your bid (up to ~£7) in return for a possible 10% increase in your overall CTR.
  • A similar trend of as SEO position 1.0 – 1.9 isn’t seen as strongly in SEO position 2.0 – 2.9, but the case of spending more for an incremental increase in CTR is still worth it.
  • SEO positions lower than 3 show a different trend than the higher SEO positions. There is a clear peak at ~£2.50, after which any more spend surprisingly shows a decrease in CTR. This is the point we were looking to find, a point after which you are actually doing worse the higher your bid. The exact values of each SEO positions PoDR is shown in the table below.
Max CPC Max CTR
1 £ 6.89 48.31 %
2 £ 6.41 44.10 %
3 £ 2.45 41.40 %
4+ £ 2.50 38.02 %

Figure 6
 
Results – Hypothesis 2
An alternative way to analyse our results is from the heatmap matrix below. This shows the PPC ranking against its SEO position, the combined CTR, CPC and an efficacy score. The score is formulated to optimise the efficiency, favouring a higher SEO position, higher CTR (which is emphasised) but lower PPC position and lower CPC. The score (K) in a simpler form is calculated below, which we seek to minimise, the lower the score, the better to be in that position. (α is a normalisation constant)
 

Figure 7

 
One reads the table from left to right by each row, so for a given SEO position we can see how the CTR, CPC and K score are distributed for each Av. PPC position. We observe that for lower SEO ranking, the K score gradually increases due to the dependence on PPC to drive the CTR.
The boxes highlighted in red are the positions with the lowest scores, therefore the most effective positions to be in. As you can see, this never occurs when PPC is ranked at exactly position 1. Even after weighting the formula to encourage a higher CTR, most low K scores are exhibited when the PPC position is lower, with the exceptions of SEO positions 1 and 6+. One could cautiously draw from this that it is not necessarily the best case to be ranking in position 1 for PPC regardless of SEO ranking.
 
Limitation/ Caveats of Project
This analysis is the beginning of what has the potential to be a bigger automated solution which we will touch on later. Currently we’ve made quite a few assumptions to simplify our models and gain quick insights.
Caveat 1: We used all the keywords to display our data because splitting up the keywords wouldn’t supply enough datapoints to draw accurate results. Each keyword has a different competitiveness hence different CPC’s, so we can’t reliably comment on the max average CPC found.
Caveat 2: We used a daily breakdown of data which has been averaged, in doing so this has diluted the data. The SERP can be a volatile environment, and the time of the day has a big influence on this but which we lose out on by averaging.
Caveat 3: There are gaps in the data where keywords weren’t ranked in a certain position, or an ad was paused on a given day, so we couldn’t incorporate them into the graphs or table.
To mitigate all the caveats above, we simply need a larger data source.
 
Next steps or future projects
As mentioned above there is a lot of potential in this study. The end goal of this project would be to create a machine learning algorithm that can guide us on what our max bid should be, based on rolling past data. We would focus on using more data, as all our caveats above were based on lack of data. We would ideally like the breakdown of the data to be hourly because the SEO rankings fluctuate throughout the day and we can we can also see the daily trends. We could also use the SEO volatility score using online sensors based on the previous 7 days, this can be incorporated into the algorithm to judge whether the max CPC is reliable enough – extended from this we could calculate the risk of using this result, and the impact to the client. Exploiting these areas would create reliable, innovative and actionable insights and be fed this straight into Google AdWords to adjust PPC bids. This will ultimately save our clients’ money.
The streamline use of Big Query ML plays a big part in the development of our accounts and inter-disciplines analysis. It’s essential to be able to use all the data possible and quickly in order to get actionable insights for our clients and make sure we are always on top of any possible optimisation. Being an integrated agency means we not only manage different disciplines but we also integrate their data. With the accessibility of machine learning solutions it is now possible to create our own models for predictive analytics.  Exploiting these solutions creates create reliable, innovative and actionable data that is fed directly into our planning as well as our strategies.
If you are interested in similar projects, please email [email protected] for more information.
Explore our full offering on our services page.