As we create more content, marketers are trying to understand the role this content plays in the buying process and which components have the greatest impact on generating conversation, consideration and ultimately consumption. So it’s no surprise that marketers are trying to understand how to leverage both marketing mix attribution and optimization models. Recently we’ve been receiving a number of questions about fractional and last-touch attribution. So we thought a brief tutorial on the topic of optimization and attribution modeling might be helpful.
There are a number of sources and tools available today to help create either model. Both attribution and optimization modeling are about improving mix and understanding the impact of marketing investments on customer behavior. Let’s begin by reviewing what these models are, when to use them, and how they are different.
Optimization Needs Attribution Optimization relies on predictive models that track non-linear relationships between specific goals and spend levels in order to predict the incremental changes in conversions based on the relationship between the variables. Many organizations attempt to optimize campaigns via A/B testing, a form of scenario analysis. Unfortunately A/B testing doesnt address the complex non-linear interactions. An algorithmic approach that simultaneously analyzes all possible scenarios is needed to see which combinations produce the best incremental results.
Attribution is based on capturing touch point data over a historical period to determine which touch points are the most effective at which stages in the buying process to support investment allocations and produce higher aggregate results. Both approaches are important for measuring and improving the performance of multi-channel, multi-touch campaigns.
Why Fractional Attribution is Good Approach Attribution is simply the ability to evaluate the performance of each touch point in the buying process. A key premise of attribution is that all touches play a role in impacting the buying process. To create any type of attribution model you need data related to converting and non-converting opportunities to develop the model. There are various approaches to attribution. Three of the most common are last-touch, equal attribution, and fractional attribution.
Last-Touch Attribution Last-Touch Attribution is based on the idea that the last touch has the greatest impact on the buying process and therefore receives the majority or all of the credit for the entire sale. Some companies take the opposite approach and use first-touch attribution which is based on the idea that the first touch is what primes the pump. Neither of these models account for all the prior or following touches that may have impacted the buying behavior. As a result you may end up eliminating important earlier or later touches because you arent sure of their value. So given some of the problems with these techniques, why do people use them? Because they are relatively easy to create.
Equal Attribution To overcome this issue some organizations use Equal Attribution, which is where equal value is assigned to every single touch. This approach assumes that all touches are equal. As a result, you may end up unnecessarily duplicating some efforts because you arent sure which touches have the greatest impact. So you may end up investing more than you need to because this approach doesnt provide insight into which touches perform best. That takes us to the concept of Fractional Attribution.
Fractional Attribution Fractional Attribution assigns a calculated weight to each marketing touch. A weight is assigned to every touch the buyer takes on their journey to purchase. Typically this weight is determined by the corresponding relative impact that particular touch will have on producing the desired business outcome, such as purchase. This approach enables marketers to take multiple prior exposures into consideration. Determining the weights requires understanding which touches perform best, which requires good data and modeling. Using fractional attribution takes understanding the statistical significance of the various touches in order to quantify their contributing effect. It is important when building this type of model and assigning weights to keep in mind that there are touches other than marketing touches that drive the desired outcome.
Which is best for you? Most attribution experts agree that fractional attribution is better than last-touch or first-touch attribution. These experts typically recommend the fractional approach where marketers assign weights to touches based on their type, and position in the buying process to create the model. Marketers can then use this model to make touch and investment decisions.
The key challenge to address with the fractional attribution approach is that buying decisions are not serial or linear; it is often a combination of touches that impact behavior. This is why some experts have created incremental attribution models, which attempts to calculate the change in revenues resulting from a particular touch. With this technique, touches are classified by the buying stage they support, and buyers are tracked as they move through the stages. Marketers use this structure to compare the effectiveness of different touches (messages and media) in moving buyers from one stage to the next in order to determine the incremental impact on cost and revenue of the different touches. As you see as we move from first or last touch to fractional attribution there is increased complexity and sophistication.
Attribution models are typically framed in terms of assigning credit for a particular purchase. As marketers we know that one touch has ripples that can affect multiple purchases and behavior. If you decide to tackle attribution, the need to combine online and offline quality data will become increasingly apparent. Attribution modeling serves as an important decision making tool. If you’re just beginning the process, take the approach of walk before your run. If you’re further along in your efforts by all means dive in.
. .
