This post discusses the Growth Marketing Attribution course as part of Track 3 (of 7) for the Growth Marketing program from the CXL Institute. The course is taught by Russell McAthy, CEO & Co-founder of Ringside Data. It covers attribution theory, strategy, and technical attribution models of Last Non-Direct Click, Last Click, Linear, Time Decay, Position-Based, and Custom (Algorithmic).
Marketing Attribution – Overview
McAthy first highlights that ‘attribution’ is one of the most overly-used buzzwords across all of marketing. It has many meanings but few people who truly understand what it is or why it’s important. More specifically, it’s not merely a math model to calculate return on ad spend or distribution of spend value. Rather, it’s about finding context to help understand consumer behavior (which of course then informs where to best spend marketing resources).
Google is the biggest player in attribution, with Google Attribution built directly into Google Analytics (and Google Adwords integrates directly with Google Attribution). Other competitors also exist, with custom (algorithmic) modeling (more on that below) being their focus. They include Visual IQ (now Neilson Attribution), Bizible, LeanData (B2B focus), and many others.
McAthy points out that 43% of companies say that measuring return on marketing investment is their top priority but less than a third of businesses are using a defined attribution model. In other words, businesses believe attribution is important but many don’t know what model to use. And, many lack the skills and talent to execute. But if you can get it right, and are willing to put in the time and budget across the whole company, attribution numbers can become the #1 decision-making tool for a brand.
Marketing Attribution Models
The course delves into high-level explanations of typical attribution models, their pros and cons, and how different businesses use them. Regardless of industry or business type, the endpoint is referred to as a ‘conversion’. While there are large differences between B2C conversions (buying a t-shirt, for example) and B2B conversions (downloading a lead-generating white paper, for example), attribution requires this end-conversion anchor point from which to work.
It’s important to note that the models are intentionally simplified to best illustrate the concept. In actuality, attribution commonly has dozens of touchpoints. The models are:
- Last Non-Direct Click Attribution
- Last Click Attribution
- Linear Attribution
- Time Decay Attribution
- Position-Based Attribution
- Custom (Algorithmic) Attribution
Last Non-Direct Click Attribution Model
In this last non-direct click model example, the third visit is inbound from a paid campaign (CPC/Search, etc.) but does not result in a conversion, and then the fourth visit is a direct visit (consumer going directly to the site) where the conversion happens. The conversion is attributed to the last non-direct click, which was the paid campaign.
This model gives heavy credit to any paid activity as the catalyst that results in the final direct visit and conversion. It is the most common model and the most limited in providing fair attribution or insights into consumer behavior.
Last Click (Last Interaction) Attribution Model
100% of the conversion value in the last click model is applied to the final interaction. It is slightly better than the last non-direct model because it is at least highly accurate. And, there is significant value in understanding the last thing that happens in a consumer journey. But this model remains stunted as it tells us very little about what led to the actual conversion.
Branded PPC campaigns (searches for a specific brand name or product) tend to do very well in last-click models because when consumer awareness of a brand is high there is a greater chance of immediate conversion. Other channels and broad searches where the consumer is still in the research phase tend to be viewed as lower value within the last click model.
Both the last click and last non-direct click models are focused on the shorter view and offer little in the way of understanding the actual consumer behavior.
Linear Attribution Model
Linear attribution provides equal value to each interaction. So if there was a research interaction, two other visits, and finally a conversion interaction, they all get equal value. The flaw here is easy enough to see – because it attributes value equally, it teaches the marketer very little about which channels and touchpoints impacted the consumer (why they interacted or which interaction impacted their decision to convert). As a decision tool, the linear attribution model offers minimal information on where to place future marketing investments.
Time Decay Attribution Model
The time decay model begins to take a more sophisticated approach to attribution. Specifically, it assigns value more heavily to touchpoints that are closest to the conversion – still redistributing value through historic visits, but scaled back over time. There are two ways to do this:
- Static Model – Typically, the last interaction gets a heavy fixed %, and the same goes for earlier touchpoints (and example would be 60, 25, 10, and 5 in the case of four touchpoints). The flaw in this static approach is that values do not change based on the amount of time that has passed between them.
- Variable Model – Takes into consideration the amount of time between each interaction and applies heavier weight to recency (the ‘decay’ in value is dynamic and it increases as time passes between touchpoints).
As with previous models, time decay falls short of providing any real consumer insights.
In the position-based model (sometimes referred to as the ‘bathtub model’), the first and last touchpoints are given heavy weighting: 40% first and 40% last, with the remaining 20% distributed between all others.
Logically, this resonates with most people as a common-sense approach. After all, the first touchpoint is where a consumer might first discover a brand, and by definition, the last touchpoint is where a consumer makes a conversion commitment. It acknowledges the importance and challenges of the first step (acquiring) and the last step (converting).
Google Analytics portrays the position-based model as the most accurate, but the fundamental flaw in the model is that the marketer is still deciding in advance what is most important rather than learning that from the consumer.
Custom Attribution Models (Algorithmic)
Custom attribution models (algorithm-based) do not rely on marketers to assign value to touchpoints without knowing for certain what is important to consumers. Instead, all data is pushed through a machine-based attribution system where complexity and number of touchpoints are not susceptible to human error or marketers manually assigning value to the wrong touchpoints.
Marketers instead provide input factors of emphasis to inform the model. For example, asking the model to focus on seasonality, or on the type of consumer devices being used.
This is by far the most costly and complicated approach to attribution modeling. Frankly, due to the platform dependencies related to custom models, it’s hard to even grasp how they work without seeing them deployed first-hand.
For companies just starting out in marketing attribution, or with limited resources, position-based models are a better starting point. But those same companies would be well-advised to begin planning now for the custom models that will soon dominate the attribution practice.