The more a user visits the site, the higher their average order value.
Now, when we look at the basic attribution models (there are more):
- First interaction: assign all conversion credit to how someone found you
- Last interaction: assign all conversion data to the last touchpoint
- Linear: assign credit equally among all touchpoints
Which model should we be using? Here’s the common problem for each of these three models with this data:
- First interaction: Ignores the 79% of revenue as that revenue occurs after the first click
- Last interaction: Ignores how a user found you initially and the fact that many users need to return multiple times to the website to convert
- Linear: Ignores the fact that the more a user returns to the site, the more the final checkout value becomes
None of these models are perfect for this data; however, we haven’t said what we’re trying to accomplish yet; and thus, we can’t answer anything.
Now, imagine your CMO looks at you and makes the statement from above:
“We have a long customer journey with many touchpoints. Our customer journey is long; however, they can’t start this long journey until they first know about us and visit our website. I want you to focus some budget in ensuring users enter our funnel. Can you accomplish that task?”
Now we have a question to answer.
We need to put some budget into the awareness stage so that users can initially find the company. In addition, these ads should be designed to get a user to enter the funnel – they are not ads designed to get a user to convert on their first visit.
Now that you know this information and directive, you can turn to your models to determine what task to take:
- Examine the top channels where users are initially finding you
- Drill down into the keywords, ads or information for that channel
- Add budget to those campaigns or sources
- Adjust the ads to be more compelling to initially visit the website and start a user understanding the product or service
To answer this question, we now turn to our analytics data to start drilling into the information. We examine the channels by first interaction data, and see that paid search is our top first interaction.
Next, we drill down into our campaigns to see where the first interaction is coming from. It turns out that most them are coming from one campaign.
So now, we can segment the data by ad group to see if it’s mostly that campaign that is leading to these conversions or if these first interaction conversions are spread among various ad groups.
Now we can see that a single ad group is responsible for the vast majority of our first click interactions.
In this ad group; we’d want to test ads to see which ones can bring in qualified visitors into the funnel so we have an idea of where to start and if there are we should pause in this ad group. So, we would examine the ad content (or ad id) for just that ad group. While you can see the actual ads within analytics, in this case, we’re going to use the creative ID to protect the advertiser data.
What we learn is that there is one ad doing a great job and another ad doing a good job of bringing in the first touch conversions:
Since this ad group is so important; we can also examine the keywords to see if there are some keywords that are more or less effective than other ones:
What we learn is that:
- 3 keywords are contributing the most to the first interaction conversions
- The first interaction ROAS varies dramatically (from 1,587% to 528%)
- The CPAs vary dramatically (from $19.88 to $6.49)
Now we can accomplish our boss’s task: Put budget into getting users to start the customer journey.
What we’d do:
- Split out the keywords based upon first click interaction ROAs into a few ad groups so we can bid and spend the budget wisely
- Create ads that are highly relevant to the keywords in the new ad groups based upon the top ads that are performing for first click attribution
- Pause the ads that are not leading to high first interaction conversions
By knowing the question, we want to answer we can use the proper model and analysis.
Most attribution questions are about bidding. However, attribution can go well beyond just bid changes to help you answer questions and take the proper action. The next time someone asks you what attribution model to use; your answer should be: I can’t answer that until you ask me a specific question. Once you know the question, then using a model to find the answer and the task to accomplish is not that difficult.
There are many more models and good questions to ask. To learn much more about attribution modeling, please join us for an upcoming webinar.
This post is guest authored by brad.geddes. Brad is a long-time industry veteran, frequent conference speaker and AdWords workshop leader. His current passion is ad testing and has co-created Adalysis a platform to let users test ads at scale and easily pull insights from their data.