To explain what the Google Analytics Model Comparison tool is, it helps to establish two things.
One: Your converting visitors will not simply stumble upon your ecommerce site. They’ve likely seen you in social media, come across a referral link to your site, seen one of your ads or found you in search.
Two: Your converting visitors will most likely have visited your site a number of times from different channels (e.g. social, referral, organic search, referral etc.) before they convert.
In fact, a common journey before conversion might look like so:
Each step in the above image illustrates a visit from a different channel before the final, direct visit to convert. Now that we’ve established that, let’s look a bit closer at the Model Comparison Tool and how you can use it.
What is the Model Comparison Tool?
First of all, what the heck is the Model Comparison Tool? Google Analytics has plenty of nice reports and features. For example, you can use it to plan content for your ecommerce site or to understand what your visitors are looking for. The Model Comparison Tool is one of GAs powerful tools for the slightly more advanced user. It looks horrendous, scary and can be quite off-putting at start. All true, but once you get the hang of it, it is mighty useful.
The Model Comparison Tool allows you to compare so-called attribution models and how they impact the value of your marketing channels. Comparing attribution gives you a better, or perhaps “truer” overview of how successful your different channels are in pushing potential clients into clients.
What is an Attribution Model?
An attribution model is how web analytics tools like Google Analytics attributes value to the channels that push converting traffic your way. This is to give you a better understanding of which channels are successful in acquiring converting traffic and where in the path to purchase the channel works the best. Simply put, the higher the value – the better it works. The traditional attribution model attributes all the value to the last channel that the converting customer interacted with before the conversion. This is the number of sales and conversions the channel closed or completed. If a converting visitor comes from a referral site before converting, ‘referral’ will be attributed the value of that purchase.
The converting visitors’ journey or path to purchase is, however, all but straight forward. Converters usually visit a site a number of times from various channels before converting.
Depending on the attribution model you use, Google Analytics will attribute the conversion value either to the first, last or a mix of the different channels in the journey. If you don’t understand the attribution, it will be difficult to plan marketing budget and digital strategy.
Google Analytics provide seven Attribution Models that you can apply and compare against each other. Here is a brief explanation of each of them.
Last interaction model
Last non-direct click
First interaction model
The linear model
Time decay
Position-based
Last interaction model
The last interaction model attributes 100% of conversion value to the last channel (regardless of whether or not it was a direct visit) with which the customer interacted before converting.
When to use?
It can be useful if your campaign is designed to attract people at the moment of purchase. If you think about the typical marketing funnel you have three main steps, awareness, consideration and decision. If your sales cycle doesn’t require a long consideration phase e.g. if you sell fast-moving consumer goods (FMCG), your clients don’t usually go through a consideration phase. In this case, the last interaction model can be useful.
The Marketing Funnel can help you understand and plan marketing strategies
When not to use?
This is the default model for most web analytics tools. It can however be misguiding since our path to purchase usually starts long before the last interaction. Remember, with this model 100% of the value is attributed to the last channel (interaction) prior to the conversion. Using this model will not give credit to other channels and can therefore cloud judgements. Let’s say your social media efforts are generating plenty of buzz and awareness around your brand but the visitors who convert need another couple of visits before they convert on your site. If we stick to the journey presented above, with the Last Interaction Model, your social media efforts will not be attributed any value and you will have a harder time to convince your boss to spend more time and effort in social.
Last Non-Direct Click
The Last Non-Direct Click model is similar to the above model but ignores direct visits. It attributes 100% of conversion value to the last channel that the customer clicked through from before buying or converting (as long as that channel wasn’t direct). This is the default attribution model in the Model Comparison Tool.
When to use?
Typing a URL directly into a browser can be an indication that the visitor is already aware of your site. If you are comfortable in that the vast majority of these visits are from customers who have already been won through a different channel – use this model.
When not to use?
This model can be a bit confusing. It is not always the case that people who type in your URL directly into the browser are aware of you in the sense that they know all about you. They might have heard about you, read about you somewhere (on or offline) and decided to type your URL into the browser.
Remember that a direct visit that ends with a conversion is commonly part of a journey with prior visits from other channels. So why attribute all the value to just the one channel before a direct visit if the visitors didn’t use that prior channel to click through?
First Interaction
The First Interaction model is the complete opposite of the Last Interaction model. It attributes 100% of the conversion value to the first channel that the customer interacted with in his or her path to purchase, instead of the last.
When to use?
This model can be appropriate if you run ads or campaigns that creates an initial awareness. Let’s say you’ve created an awareness campaign in social media. After the campaign you want to see how successful your social efforts were. With the Last Interaction and Last Non-Direct models you would most likely get a very low value attributed to the ads used in creating awareness. Again, looking at the marketing funnel, awareness is at the very beginning of the path to purchase. People don’t usually buy a product the same day they learn about it, but sniffs around a bit, read reviews, compares similar products etc. and then converts. During this process they are likely to visit your site again and again via different channels. Since your social efforts are at the beginning of that journey, the First Interaction model will attribute value to these interactions.
When not to use?
Remember the issues with the Last Interaction model? The same applies to this model. The first interaction with a converting visitor might have been amazing and jaw-dropping but the converting visitor still needed a couple more visits to your site to be convinced. Without those other visits, the conversion would probably not have happened.
Linear Attribution
This model is nice but perhaps a bit too nice. Instead of giving the first or last interaction all the credit, everyone gets equal amounts of credit to each channel interaction on the way to conversion.
When to use?
The Linear Attribution model can be useful in e.g. B2B where it’s common to have longer sales cycles. Throughout a longer sales cycle you’re typically using a number of channels to maintain contact with potential customers. Therefore you want to attribute value across all of them. Simply giving value to the last one will ignore all the earlier efforts.
When not to use?
With this model everyone gets equal amounts of cred, but it won’t give you too much insights into which channels really push your potential customers to become customers if you have a shorter sales cycle.
Time Decay
As the name applies, the Time Decay model, attributes more value to the channels the closer they get to conversion. The channel closest to conversion will be attributed more than those in the beginning of the journey. The Time Decay icon explains it better than me…
When to use?
This model is, in my view, one of the better general models that you can use to get a good overview. It makes sense in that all interactions should be attributed value, but not equal amounts. The first interactions didn’t make your converters convert, but it got their attention nothing more, nothing less. The channels used closer the conversion helped to convince them to actually convert. Therefore they are in a way more important.
When not to use?
If you believe that the first interaction is the most important one as in the case of a brand awareness campaign, this model can be misguiding. Also, if you strongly believe that last click shouldn’t be attributed to a direct visits, this model might not be for you.
Position-Based
The Position-Based model gives percentage of credits to each step in the conversion path based on its position – first, last or in-between.
When to use?
If you felt that the Linear Model was a bit too lukewarm, this one is for you. It attributes value to all the channels but more so to the first and last. Both the first and last interactions are special in that they kick of the journey and then completes it in the other end. Still, this model takes the other channels in consideration since they also played a part, albeit (possibly) a less important part.
When not to use?
I like this model, I really do, but it’s not for everyone. It’s not the kind of model you start out with since it has an uneven (position-based) value distribution. If you don’t feel comfortable explaining an attribution model in general, it can be tough to motivate why you’ve chosen this one when you’re interrogated about your data insights and action plan.
Last AdWords Click
To group all AdWords ads into one channel doesn’t give your hard work and 50+ ads justice. With the Last AdWords Click model however, you attribute 100% of the conversion value to the ads that were most recently clicked by the converting visitors.
When to use?
With the Last AdWords model you can identify and attribute credit to the AdWords ads that closed the most conversions. If you feel/think that the last AdWords ad that has been clicked before conversion, is the one that most accurately tells you which are the most successful in sending converting traffic to your site, this is the one to use.
When not to use?
For the same reasons you shouldn’t always use the Last Interaction model, this model can give you skewed numbers in that it’s too simplistic.
Tool
Let’s have a look at the actual tool in its totality. With the Model Comparison Tool you can compare up to three models at a time. As you do, you should keep an eye out for those that change significantly from model to model. Before I explain that further, let’s look at the different bits and bobs of the tool.
At the top of the Model Comparison Tool you have a couple of settings. These will affect the data that is being pulled into the actual comparison tool.
Conversion: By default, this is set to ‘All’ which means that you are not only tracking ecommerce conversion but also goal conversions. In Google Analytics, ‘conversion’ refers to any kind of conversion whether that is a purchase or a goal conversion. I find it confusing to look at ‘All’ the conversions. I usually either look at one specific goal or ecommerce transactions only.
Type: By default this is set to ‘All’ which means that it checks conversion through all channels. If you click ‘AdWords’ it will show you conversions through AdWords only which can be handy if you want to understand how your AdWords campaigns have performed.
Look-Back Window: the conversion journey starts at some point. With the Look-Back Window you can decide from which point you start considering the path to purchase. By default it’s set to a 30 day look-back.
Below these settings you can chose which models to compare. In the below example there is only the Last Interaction model, which is the default model. By clicking the drop-down arrow in the ‘Select model’ you can pick which one(s) to compare with. You can also change the default model by clicking the drop-down arrow next to the Last Interaction icon.
Below these setting is the actual model comparison tool. In the below example I’ve only used the last interaction model. I’m not actually comparing anything. With just the one model applied you can see the number of Last Interaction Conversions i.e. the number of conversions from each channel, and Last Interaction Conversion Value i.e. the total value generated per channel.
The Model Comparison Tool with one Attribution Mode
When I pull in two more attribution models to compare with, the comparison tool will look like so:
Model Comparison Tool with three Attribution Models
As you can see in the above comparison, I’ve compared the default Last Interaction model with Linear and First Interaction models. Yes, there are a lot of numbers here but the interesting bit is on the right side of the table. The two last columns to your right are a comparison in percentage from the Last Interaction model to the ones I’ve added. The default setting is comparing the ‘% change in conversion’. You can change this to ‘% change in conversion value’ if that suits you better. Depending on what you sell on your site, the change in conversion value might be more interesting.
The arrows in the columns to your right indicate a positive or negative change compared to the Last Interaction attribution model. If you look at both Organic Search and Email, we see quite a change both in the Linear (and even more so) in the First Interaction model. This means that Organic Search and Email are more valuable in (especially) the First Interaction model i.e. Organic Search and Email are stronger in the earlier stages of the path to purchase. They tend to start converters on their journey. If you look at Referral, you’ll see that it’s quite the opposite. Referral traffic is more often used as the last interaction before a conversion.
Now, there is more to it than looking at the arrows pointing up or down. Attribution modelling is fairly advanced stuff. You have to consider the type of industry you are in and what your typical sales cycle looks like. You also need to consider the other numbers in the above table. If you look at Organic Search and Referral for example, we can see that Organic Search has produced a much higher number of conversion and conversion value. Although it might be tempting to go and change your referral strategy, it might have little impact compared to other more high value channels.
Take away
The take away is simply to start playing around with the Model Comparison Tool. Understand your typical life cycle and what a customer journey looks like and apply different Attribution Models to see if you can spot huge differences between the models.
As with everything in web analytics, it’s important to remember that you need a fair bit of empathy and imagination. What is the main reason to why you don’t convert on your first visit to a site? Why did you use channel x before you converted?
At LogicSpot we drive our decisions based on data analysis. We don’t let hunches drive our clients business. If you want to talk about Attribution Modelling or anything related to web analytics, please don’t be a stranger.
