Once you start diving into trying to understand how well your marketing channels and campaigns are working, you quickly stumble into the mysterious world of attribution models. Various attribution models have been created to help marketers decide which channels (e.g., paid search, organic search, email, etc) and specific campaigns within these channels are responsible for generating conversions. These models are supposed to show you what is working and what is not. But it’s not as simple as that. What does “working” mean when you’re presented with five, six, or more attribution models that show vastly diverging data? And it’s all complicated by the fact that most user paths to conversion include many interactions (clicks) originating from different channels and campaigns. Let’s find out what attribution model works for you!
Here we’ll look at the popular attribution models that have been around for some time, and at a new one that uses machine learning (ML), a form of artificial intelligence (AI), to give you a dynamic “custom” model that builds a more accurate picture of what is actually happening on your site. First, let’s break these models into two classifications: single-touch and multi-touch.
Single Touch Explained
Single touch looks at the first or the last touchpoint (an interaction on your site, usually a click) and gives credit for (attributes) each conversion (In the e-commerce world that is typically a purchase) to that touchpoint..
The First Touch model credits a conversion to the first interaction made by each user. This is not necessarily the touch that resulted in the conversion since the user could have made a purchase from a subsequent touchpoint or a touchpoint during a subsequent visit. So, in the first touch world, credit for a user who comes to the site for the first time from a paid search ad will be given to paid search, even though the user completed their purchase from a subsequent touchpoint, for example from an organic search (the last touch). So, First Touch shows how visitors who completed a purchase first came to your site.
In the Last Touch model, the purchase is credited to the last user interaction, in the example above, Organic Search. There is a refinement of the Last Touch model and that’s Last Non-Direct Touch. So, let’s say in the above example there were three touchpoints: Paid Search, Organic Search, and Direct (where a user had bookmarked or entered your domain directly into the search bar). Last Non-Direct Touch would overlook the Direct touch and give credit to Organic Search. The non-direct model only takes into account those touchpoints you can directly impact from your own marketing efforts. You have minimal immediate influence on Direct interactions (as Direct clicks are typically originating from users who have remembered or bookmarked your URL). Eliminating Direct clicks, except when there is no other interaction, gives you a better picture of how well the campaigns you’re managing influence conversions.
Most customer journeys to purchase are based on a large number of touchpoints, depending on the site this could be eight or more touchpoints leading to purchase. Each of these touchpoints makes some contribution to the purchase.
So the multi-touch models give each touchpoint some of the total credit. Each of the multi-touch attribution models makes a different calculation of how credit is shared based on an assumption of what each touchpoint is worth. To give the person analyzing the data a rough idea of the relative worth of each touchpoint, each of the multi-touch models makes what are simplified assumptions to portray the value of each touchpoint along the conversion path.
Here are the popular multi-touch models:
Linear attribution: gives each touchpoint equal credit, so in the case of a user journey to purchase that had 5 touchpoints leading to a purchase, each touchpoint in the Linear model would get 20% of the credit.
Linear Non-Direct Attribution
Linear Non-Direct Attribution: same as Linear except Direct touchpoints are eliminated. Again, this model only considers the touchpoints you can influence through your marketing efforts.
Position Based Attribution
Position Based Attribution: gives the First and Last touchpoint 40% of the credit each and divides up the other 20% equally among the intervening touchpoints.
Time Decay Attribution
Time Decay Attribution: gives the highest credit to the last touchpoint, a bit less to the next to the preceding one, and progressively declining credit to each of the preceding ones.
Because these are heuristic (best guess/rule of thumb) models they don’t account for user behavior on a specific site. User behavior varies from site to site. These heuristic models cannot account for the unique characteristics of visitors to and products on a site. Because these heuristic models are a simplistic way of attributing conversions to multi-touch scenarios, they don’t reflect the true weight or importance of a channel or a campaign in generating revenue. The fact is that certain channels/campaigns are more important than others and in a multi-touch scenario some touchpoints are not important for some conversions (which likely would have happened without a marginal touchpoint).
For this reason, some marketers create Custom Multi-Touch Attribution models that they think more accurately reflect the distribution of credit in a multi-touch scenario. But creating a custom model and maintaining/adjusting it over time is a very significant job even at the channel level. Doing it for campaigns that are constantly being revised becomes a herculean effort.
Machine Learning (ML), a branch of Artificial Intelligence, is ideally suited to the “heavy lifting” needed to analyze the vast amount of data being collected by an attribution platform. One of ML’s strengths is its ability to “digest” huge amounts of data and recognize the patterns in the data. This is the base that makes it possible to automate the creation of attribution models that reflect all the many user journeys taking place on a site. Machine Learning can also determine the importance of each touchpoint leading to conversions in a multi-touch scenario. Rather than the simplified attribution in the standard heuristic multi-touch model, ML can run scenarios to test how important each touchpoint was. Since every site is different, a Machine Learning Attribution model can actually generate a weighting of each channel or campaign unique to each site.
ML also opens up analysis that extends the functionality of an attribution platform. In a future blog, we’ll explore some groundbreaking enhancements AdAmplify is adding to its Dimensions Attribution platform.
Which Single Touch Attribution Model Should I Use?
The answer depends on what you’re trying to understand. First Touch gives you a picture of how well each channel or campaign is doing in bringing new customers to your site. This is important in understanding where you should focus your “recruitment” tactics for new visitors who are likely to convert. AdAmplify’s Dimensions Attribution adds an additional nuance by also showing you the subsequent revenue generated by each customer. That way you get to see the performance of the channel/campaign in recruiting first-time purchasers and also how much these new customers make in subsequent purchases over a selected timeframe.
Last Touch and Last Non-Direct Touch show you the ability of each channel or campaign to convert. This is important in understanding where you should focus your efforts to turn visitors into paying customers. Last Non-Direct Touch, in eliminating Direct touchpoints, lets you focus on the channels and campaigns that you can control (again, you have little control over Direct touchpoints). Again, Dimensions Attribution separates initial and subsequent purchases so you get a better overall picture of a channel’s or campaign’s performance, first in driving first purchases and then subsequent purchases. This helps you to understand the performance of any channel or campaign when it comes to first or subsequent purchases. For example, email may be a small contributor to first purchases (as the user may not subscribe until they make a purchase), while subsequent purchases are typically a major contributor to subsequent purchases for many companies.f
Which Multi-Touch Attribution Model Should I Use?
While the Single-Touch models illustrate the touchpoints most important in recruiting new customers or in converting directly to a purchase, multi-touch models offer a needed representation of the multiple touchpoints that lead to conversions on a site. The heuristic attribution models such as Linear and Position-Based are a “best guess” representation. Machine Learning though is an accurate data-based analysis that uses all its strengths to present an accurate attribution picture unique to each site. It also provides the user with a dynamic picture, constantly recalculating the contribution each channel or campaign is making. For Dimensions, this analysis is updated daily. So, the answer is to go with the attribution model that has a data-based foundation, not a “best guess” one.
Attribution Models Compared
To see the pros and cons of the different models, along with a call out on some of the enhancements AdAmplify has made to the standard models click the link below.