ATTRIBUTION MODELING
ATTRIBUTION ACROSS DIGITAL CHANNELS
BACKGROUND
Ever since the first company spent their first dollar to advertise their products on the internet, the ever so simple question of “how much money am I going to get back on my investment” arose. Enter the age of online reporting and analytics, which depended upon 3rd party ad serving technology to provide conversion metrics. These conversion metrics, although insightful, were anything but perfect, but online marketers didn’t have anything else to work with at the time and were grateful for what they were able to receive.
But what was the problem with this conversion data? The problem was that the conversions were all tracked using “last click” interaction as the sole aspect of the user’s purchase cycle that was being given conversion credit. Last interaction is only one part of the story. As important as this concept is, it is also easily overlooked.
Why does the whole story matter? Today, consumers are faced with more stimuli than ever before, leading to make the decision making process a more complex one. A prospect can experience multiple touch points or interactions that eventually lead up to that “last click”. So measuring the last click is like knowing the ending of a book, but completely missing the body and introduction. So how do we attribute credit to each interaction leading up to a conversion? That’s the Attribution Challenge we are about to discuss.
PART I
The Attribution Challenge
In the world of Web Analytics, we are faced with multiple concepts that can be complicated, such as statistical modeling, big data, and others. One key aspect is attribution modeling. When people hear the word “attribution”, they run away screaming. Add “modeling” to it and they’ll jump off a cliff!
So what is attribution? In short, it’s a set of rules that define how to give credit to the different interactions within the path to conversion.
There are tons of software companies and gurus out there that claim to have or know the best solution to the attribution challenge. The issue lies around the vast number of unknowns we have to consider and, to top it off, there is a lot of info that we can’t even capture. To overcome this overwhelming task, the intent of this article is to break down the attribution challenge in smaller pieces to try to understand it.
Attribution might sound daunting, but we promise at the end of this rainbow waiting for you is a pot of gold – literally. Add the following basic fundamentals of attribution to your toolkit and soon you’ll have a deeper understanding of attribution across digital channels, giving your company a true chance to get closer to your fiscal goals.
First off, let’s stop referring to attribution as a single entity. In fact, we can define four distinct attribution challenges:
- Attribution across Digital Channels, which covers the challenge of attributing the impact of marketing across digital marketing channels (Paid Search, Display Media, Social Media, Email, etc.)
- Attribution across Multiple Screens, which covers the challenge of attributing the impact of marketing across multiple devices (mobile, desktop, tablet, etc.)
- Attribution for Offline to Online Conversions, which covers the challenge of attributing the impact of offline marketing on online conversions
- Attribution for Online to Offline Conversions, which covers the challenge of attributing the impact of online marketing on offline conversions
Attribution evidently comes in various shapes and forms, but let’s focus on Attribution across Digital Channels for now.
To understand attribution, let’s start with the example below, where our dollars were invested in Organic Search, Paid Search, Social Network, and Direct and received 31 conversions:
The question remains, how much credit do we give to each of the steps in the conversion path? Most web analytics tools will give credit to the direct channel because it was the last click prior to the conversion. This is definitely not the right way to look at conversions and is a prime example of why attribution and the concepts of different attribution modeling have become such an important aspect of measuring and telling the story of a company’s digital marketing performance.
Most analytics and attribution platforms provide a way to compare attribution models other than last click/last interaction. With that said, let’s move on to our next section, attribution model comparison.
PART II
Multi-Channel Attribution Models
Model Comparison
Most analytics and attribution software offer features that allow you to compare multiple attribution models at the same time. This allows you to easily view and attribute credit to all marketing channels that led up to a conversion, whether it was a soft (email sign up, brochure request, etc.) or a hard conversion (an actual purchase) and compare the conversion attribution / distribution based on different models.
Below, you can see a comparison between the Last Interaction Model and the Linear Model:
If you look at the column on the right, the arrows will give you an initial indication on how to shift your budgets. But wait! Not so fast, this will only work if you apply the correct model. If your model is flawed, the recommendations you’ll receive will be flawed as well. Now, let’s take a look at most of the attribution models utilized today:
Let’s use the following cross channel conversion example as we review the different attribution models below:
Cross Channel Conversion Path example:
Last Interaction/Last Click Attribution model
This is the model most often applied to performance and attribution reports. As you can see, this model makes no sense, yet it was the only model available to marketers online for many years. This model gives credit for all 31 conversions to the direct channel. The other channels (organic, paid search, and social network) were also involved, they should get part of the credit as well for influencing the conversion.
First Interaction/First Click Attribution Model
This is the opposite of the last click / last interaction model. It gives all of the credit to the first interaction / click. Based on our example, this model would give credit to organic search. It’s like giving full credit to my first job for getting my current job. I needed jobs 2, 3, and so on to get to my current one. On the perspective of this model, jobs 2, 3, etc. were meaningless.
Linear Attribution Model
Things start to get better. This model gives equal credit to all channels. However, why would you give equal credit when you know life is not that perfect and more specifically some channels are better than others at driving interactions and conversions? There are definitely other models that should help you get a better idea of which channels provide more value and which ones don’t for your company scenario.
Position Based Attribution Model
This model, by default, gives 40% credit to both, the first and the last interactions, and the remaining 20% is then distributed evenly among the interactions that occurred in the middle. This model is our favorite thus far, but we recommend that you don’t utilize the default values and rather give more credit to the last interaction, because it ultimately gave the final nudge to the consumer to make the conversion. At the end, you have to experiment to determine what works best.
Time Decay Attribution Model
This just keeps getting better and better. The purpose of this model is to give more credit the closer we get to the actual conversion. By looking at the graphic for this model, you can immediately see that it makes more sense. An argument can be made as to how much credit to give to the last touch points versus the previous touch points, however, it makes complete sense that the farther away a channel is from the conversion, the less credit it should receive. This is definitely a good place to start your attribution model.
One out of the seven models you can use with a certain degree of confidence is the Time Decay Model. The Position Based Model isn’t optimal, but can help you test different scenarios.
After you have experimented enough with the Time Decay Model, you can start creating a customized attribution model that may suit your business needs more closely.
PART III
Introduction to Custom Attribution Models
We recommend you don’t create a custom model until you fully understand the Time Decay Model and its implications.
To create your custom attribution model, you might want to start with one of the other models that closest fits your needs as your baseline. Then, you can modify it based on the factors that you deem important to be part of your model. Among the questions that need to be asked about your business are:
- What user behavior is important to you?
- What booking windows do you need to consider?
- Are there any soft conversions that you need bring into the mix and that have economic value? (e.g.: email signups, chat requests, etc.)
Answering these questions will give you important context when making decisions in regards to the custom attribution model.
There are a series of steps that you need to take in order to create your custom model. For instance, if you are using the Position Based Model as your baseline, then these are the steps that you will need to take:
- Specify the amount of conversion credit based on the position
- Select the lookback window
- Select the engagement based on credit option
- Apply custom credit rules
Obviously there are going to be different steps for each model, which are ultimately based on each company’s preferences.
Back in the day, “last click attribution” was the only way for marketers to measure success, and we stuck with that methodology for many years until tracking and attribution technology started to evolve. Today, advances in attribution modeling are allowing us to improve our understanding of the path to conversion across different digital channels. Last click attribution needs to become a thing of the past and we need to welcome different models such as Time Decay and Position Based in order to have a better understanding how each component of the digital marketing media mix contributes to a conversion.
In a future article, we will discuss the challenges of cross-device / cross environment attribution.
