Attribution Models using Google Analytics
I’ve often read that attribution models are not the panacea for advertising budget allocations, nor is it a perfect science for programmatic predictions and deterministic models. Every source that I have found, claiming attribution isn’t a panacea, do not provide methods for attribution models using Google Analytics. Attribution models can be a vital part of analytics for insights that lead to budget refining and for improving campaign performance with mathematical precision. Consequently I’ve found when using attribution models the analyst can further identify where the campaigns need to be improved from the source, and also the medium. Analysts also identify where key changes are required to an advertising campaign’s call-to-action, the offer, copy and content, and UX. Through the fine tuning of the attribution models it is possible to determine the effect of an increased and decreased budget — by channel. Finally, when everything is correctly linked, coded, and setup, an analyst can just as much provide the same level of predictions to online advertising as much as combinations of online with offline advertising.
The weakness of attribution modeling, if there is one, isn’t in the modeling, it’s more a weakness in the methodology applied to the modeling setup. Providing optimized performance cannot start with attribution models, we have to earn our way into it. An evolutionary process if you will. The starting place for the evolution when using Google Analytics (GA) is data integrity (from a previous article) and consumer Acquisition. Form the acqusition opportunities as they are provided in the GA data management API we can identify where the website traffic is coming from. The acquisition is primarily divided into channels, 8 channels as they are recorded in the out-of-the box version of GA.
- Paid Search
Within the Administration (Admin) opportunities, as provided in GA, we can connect AdWords, Doubleclick, Adsense, Youtube Search Console, and a few other Google API resources. These additional API connections identify even more detail from the acquisition channels. This acquisition (‘A’) section of GA provides an analyst with the summary understanding of where traffic/users are coming from and provides the foundations for insight to why they are coming. The next section of the GA API opportunities reveals what those acquired users are doing on the website.
BehaviorsIn this ‘B’ section of the GA API we can better determine what the users are doing, what are they engaging with, and gain insight to how well the acquisition of traffic is relating to the design of the website. This combination, we often call the user experience or UX for short, provides the analyst with answers for 2 primary questions:
- How well does the website meet the user’s expectation
- Are the acquisition methods delivering the right users
The basic, out-of-the-box GA opportunities help an analyst discover the entry pages, exit pages, the time a user spends on site, and other general non-sophisticated click-tracking metrics. To get more advanced analysis therefore requires a few additional settings that are dependent on the client needs can be made within the admin console. Also, and depending on the website tracking needs we will use Google Tag Manager (GTM) in combination with GA. The consequential result from using these more advanced techniques are essentially the analyst can define events and custom dimensions that are key to identifying deeper insight for answering questions 1 and 2.
The level of sophistication and exact measurement protocol, data integrity and governance, as well as the specific tags, scripts, variables, conversions/goals, funnels, etc. are provided through a tracking assessment guide. A website tracking assessment guide (STAG) is a document prepared by a highly skilled Google Analytics integration professional. With this keen understanding and guidance the next logical and obvious area the analyst requires knowledge of are conversions. Since the STAG is critical to the measurement strategy it will be the topic of coming article.
Conversion analysis requires the analyst to apply dollar values to each of the goals. For an ecommerce website, and for websites that provide direct selling opportunities, setting a dollar value to the conversion is obvious. However, even websites that do not provide a users with a direct purchasing opportunity such as sites that are designed to generate leads, or drive phone calls to a call center require the analyst to set a dollar value. Even the value of a click, engagement, and goal from sites where the businesses have a long buying cycle or non-direct buying cycle such as medical, banking, insurance, recovery, real estate, etc. the analyst loses key insights to elements of the conversion analysis. In all verticals, and in every case where attribution models are desired, a dollar value for each goal is necessary.
Core Principals of Attribution Models
As we can readily identify, due-to progressing through the ABCs of GA, the more we drill down in the analysis to deeper levels, in addition to digital and web analytics we also discover the need for understanding several key business principals. These core principals require an advanced understanding of:
- Statistical Analysis
- Statistical Modeling
- Business Intelligence
At the bottom of the C-section of the GA API we find attribution modeling. Here we discover the pinnacle of what GA can provide the analysts. Once the attribution models have been developed, and are providing the richest possible insights to revenue and cost cutting, we move forward with other, more sophisticated analysis platforms such as Analytics Canvas, Big Query, SAS, MiniTab, etc. But, before we do, let’s back up a step or two and take a closer look at preparations for attribution models.
Before defining the attribution models that are best for a client’s business there are many considerations and prerequisites to consider. Such as determining the website structure, its taxonomy and content which requires a thorough understanding of the target consumer, stages of the buying journey, and the content that will be needed to meet the consumer at each stage of the journey. Further more, we must understand what causes the consumer to buy, what choices determine which competitor the consumer will buy from, which consumer segments are right for the client and which ones are not.
Once defined, these segments can further lead to the analyst defining the advertising strategy and channels that reach these key segments. The website, landing page content, and engagement that each targeted segment expects are prepared. Knowing how to communicate with each segment, developing the communication, and providing the exact next-steps for the consumer brings us to a point in the advertising process where we need metrics. We have to determine what the key performance indicators are for measuring the advertising segment, content, and engagement process. When we are able to measure the channels, and landing pages, and events, both individually and in combinations we can make decisions from the insights gained to optimize the dollars invested to the dollars gained; maximize the ROAS.
Key Metrics and Segment
The attribution model at this stage of the advertising implementation requires the analysts to provide two metrics, and additionally the analyst will need to define campaign segments in the GA API.
- The cost for each medium
- The revenue from each medium
- Campaign specific segments
The C-section in the GA API requires that the analyst define and provide the values for each of the two metrics. We provide the values from the Administration console. Additionally the analysts will require the use of GTM to identify custom dimensions for user ID, Session ID, time-stamping. GTM is also necessary so that we can apply advanced techniques by defining data-layer variables and java-scripting that enhance the tracking of the consumer through site visit, engagement, and each defined conversion step.
It’s Probably Worthwhile,p>At this point you may be thinking attribution modeling is too sophisticated, too time consuming and too expensive. And yes, I will agree that at this level of sophistication the average web or digital analyst probably does not have the time nor the skill available to accomplish attribution models. Consequently they are not going provide the analysis, and insights. However, the benefits are tremendous. Especially consider the value from answers to these :
- Which channel is providing the highest ROAS (return on ad spend)
- Which consumer segments are key to the acquisition process
- Which content is failing to move the segment to the next steps
- What would happen if we increase spending in TV, radio, online channels and mediums
- What are we losing with the current strategy, what can we gain if it changes
- What is the impact on pricing changes, coupons, upsales
We have to!
We need this.
Evolution of Attribution Steps
The steps in the scientific evolution from the ABCs of GA to the attribution modeling:
- Data Integrity
- Journey Mapping
- Site Taxonomy
- Site Tracking
- Revenue Tracking
- Campaign Tracking
- Cost Tracking
- Statistical Modeling
- Predictive Modeling
In conclusion, If you decide to learn and apply attribution models yourself, you will have to take on the cutting edge definitions and creativity in the evolution. If you are going to hire an agency to build the models for you, you will have to get comfortable with — trial, test and change, repeat – to create the right models for your business. At a minimum, you will need to hire an agency with a skilled statistician who can understand the 8 core elements to keep every step of the evolution focused and on track.
The good news, the silver lining to the emerging technology, is that when you follow the 10 steps of the process from Data Integrity to Journey Mapping, etc. and allow the process to naturally evolve, earning your way along through each step, the panacea for optimizing the advertising ROI can be realized. At the very least you will have realized higher returns and you will have the right data for moving forward to big data applications such as Analytics Canvas, Big Query, SAS, or Minitab.