Email Marketing Conversion Attribution Case Study
Posted: Tue Dec 03, 2024 9:09 am
In this post, we are going to show, through a real case, how different the results (in conversions) attributable to the email channel can be depending on the attribution model we use. To illustrate this, we have taken as a reference the conversions in the period from June 1 to July 7 as shown by Google Analytics and the ESP.
summary_models
LAST INTERACTION
According to this model, there were 820 conversions that occurred after clicking on the email. In other words, there were 820 users who converted directly namibia business email list from the email without any other channel intervening until the conversion.
The advantage of a model like this is that it gives us a very accurate view of the conversions that are most certainly attributable to the email, however, it is true that it is not excessively restrictive as we will see in the following models.
FIRST INTERACTION
Another different point of view is that in which we take into consideration the weight that receiving an email, reading the subject, opening it and clicking on it have throughout the user's customer journey. In other words, does the interaction with the email influence the user in such a way that it conditions (to a greater or lesser degree) a subsequent purchase process? It is reasonable to think so. For example, it may happen that the user who has received an email reads the subject and for whatever reason does not open it. If we have been able to transmit valuable information to him, for example the start of sales, it is likely that he will go "directly" to the e-commerce when he is ready to browse; or he may see an advertisement for our brand and, conditioned by the email he received previously, decide to interact with this advertisement. In cases like this, it is evident that the email can affect other interactions that culminate in a conversion (we understand that email, due to its "push" nature, responds well to behaviors like the one we have just described). Models of this type are called “first interaction” models, since 100% of the conversion value is attributed to the first channel with which the customer has interacted. According to the example we are discussing, we would find that there have been 1,275 conversions attributable to email. In this model, no weight is attributed to email when it is at an intermediate point in the customer journey.
First Interaction
First Interaction
LINEAR MODEL
In the linear model, the same percentage is attributed to the different channels that have intervened in the path to conversion. For example, if a user has interacted with a banner and visited the site and has not converted, two days later he receives an email and accesses the site again through it without converting, and three days later he searches for the brand on Google, accesses the site and converts, according to this model the banner, the email and Google will share the sale in the same proportion (one third each). According to this model, there were 1,003 conversions attributable to the email.
Linear model
Linear model
TIME DECAY
Google explains that this model “…is based on the concept of exponential decay and primarily values touchpoints closest to the time of conversion. By default, the Time Decay model has a half-value duration of seven days, meaning that a touchpoint that occurs seven days before a conversion receives half the value of one that occurs on the same day as the conversion. Similarly, a touchpoint that occurs 14 days before receives a quarter of the value of one that occurs on the day of the conversion. Exponential decay extends based on the lookback window, which defaults to 30 days .” In other words, the closer the email was to the time of conversion, the more weighted it will be in conversions. According to this model, there were 1,018 conversions.
Deterioration over time
Deterioration over time
BY POSITION
This model combines the first and last interaction models. Instead of attributing conversions to the email based on whether it is the first or last interaction, it is split between both. A common example is assigning 40% of the credit to the first and last interactions, and 20% to the interactions in between. According to this model, 1,033 conversions were attributed to the email.
In this case, conversions that come from direct traffic are not attributed to direct traffic but to the last indirect interaction. For example, let’s suppose that a user accesses the site after interacting with an email but does not convert. Two days later, they access the site directly and make a purchase. This model will attribute the conversion to
. If we compare the conversions according to the last interaction with those of the last indirect click, we find that in the second model they are almost double (820 vs 1,514). Our interpretation of this data is that email is a very valid channel for communication (branding).
Last indirect click
Last indirect click
ESPs typically have a cookie-based tracking system through which conversions are assigned to a campaign when they occur within 30 days. This includes all variants, from users who clicked on the email and made a purchase, to others who clicked but converted days later after accessing the site directly or indirectly through some other channel.
Not surprisingly, this is the system in which the highest number of conversions appear, specifically 1,908.
We consider that the models based on the last direct click and the first interaction are fairly balanced. Attributing all conversions in which the email has participated, either directly or indirectly, as ESPs do, may be excessive (due to excess), while the last interaction does not consider a real impact that the email does have along the path to conversion. Which model do you consider best?
summary_models
LAST INTERACTION
According to this model, there were 820 conversions that occurred after clicking on the email. In other words, there were 820 users who converted directly namibia business email list from the email without any other channel intervening until the conversion.
The advantage of a model like this is that it gives us a very accurate view of the conversions that are most certainly attributable to the email, however, it is true that it is not excessively restrictive as we will see in the following models.
FIRST INTERACTION
Another different point of view is that in which we take into consideration the weight that receiving an email, reading the subject, opening it and clicking on it have throughout the user's customer journey. In other words, does the interaction with the email influence the user in such a way that it conditions (to a greater or lesser degree) a subsequent purchase process? It is reasonable to think so. For example, it may happen that the user who has received an email reads the subject and for whatever reason does not open it. If we have been able to transmit valuable information to him, for example the start of sales, it is likely that he will go "directly" to the e-commerce when he is ready to browse; or he may see an advertisement for our brand and, conditioned by the email he received previously, decide to interact with this advertisement. In cases like this, it is evident that the email can affect other interactions that culminate in a conversion (we understand that email, due to its "push" nature, responds well to behaviors like the one we have just described). Models of this type are called “first interaction” models, since 100% of the conversion value is attributed to the first channel with which the customer has interacted. According to the example we are discussing, we would find that there have been 1,275 conversions attributable to email. In this model, no weight is attributed to email when it is at an intermediate point in the customer journey.
First Interaction
First Interaction
LINEAR MODEL
In the linear model, the same percentage is attributed to the different channels that have intervened in the path to conversion. For example, if a user has interacted with a banner and visited the site and has not converted, two days later he receives an email and accesses the site again through it without converting, and three days later he searches for the brand on Google, accesses the site and converts, according to this model the banner, the email and Google will share the sale in the same proportion (one third each). According to this model, there were 1,003 conversions attributable to the email.
Linear model
Linear model
TIME DECAY
Google explains that this model “…is based on the concept of exponential decay and primarily values touchpoints closest to the time of conversion. By default, the Time Decay model has a half-value duration of seven days, meaning that a touchpoint that occurs seven days before a conversion receives half the value of one that occurs on the same day as the conversion. Similarly, a touchpoint that occurs 14 days before receives a quarter of the value of one that occurs on the day of the conversion. Exponential decay extends based on the lookback window, which defaults to 30 days .” In other words, the closer the email was to the time of conversion, the more weighted it will be in conversions. According to this model, there were 1,018 conversions.
Deterioration over time
Deterioration over time
BY POSITION
This model combines the first and last interaction models. Instead of attributing conversions to the email based on whether it is the first or last interaction, it is split between both. A common example is assigning 40% of the credit to the first and last interactions, and 20% to the interactions in between. According to this model, 1,033 conversions were attributed to the email.
In this case, conversions that come from direct traffic are not attributed to direct traffic but to the last indirect interaction. For example, let’s suppose that a user accesses the site after interacting with an email but does not convert. Two days later, they access the site directly and make a purchase. This model will attribute the conversion to
. If we compare the conversions according to the last interaction with those of the last indirect click, we find that in the second model they are almost double (820 vs 1,514). Our interpretation of this data is that email is a very valid channel for communication (branding).
Last indirect click
Last indirect click
ESPs typically have a cookie-based tracking system through which conversions are assigned to a campaign when they occur within 30 days. This includes all variants, from users who clicked on the email and made a purchase, to others who clicked but converted days later after accessing the site directly or indirectly through some other channel.
Not surprisingly, this is the system in which the highest number of conversions appear, specifically 1,908.
We consider that the models based on the last direct click and the first interaction are fairly balanced. Attributing all conversions in which the email has participated, either directly or indirectly, as ESPs do, may be excessive (due to excess), while the last interaction does not consider a real impact that the email does have along the path to conversion. Which model do you consider best?