But can you accurately identify the real value of each of the media in your media mix ? Would you be able to say with certainty which of the media could be turned off without impacting the entire chain of your media mix? This is a decision that, if based on a functional attribution model, can be simple, but the big question is whether you have enough confidence in your plan. Let's analyze this question based on a real case in which Math Marketing worked with one of the largest fintechs in Brazil.
Real case of media attribution
The detailed scenario of this case occurred in a fintech company where there was a first-click and last-click vision that depended on a large investment in paid media and audience purchasing. There was no complete vision of the purchasing journey, in the words of the client itself: “Campaign tests are done somewhat blindly…” Math Marketing had the challenge of creating a model that provided transparency, as well as the measurement of real investment results and insights into possible ways to improve results. To do so, we initially applied our framework created by our data scientists.
Media attribution framework
Deep Dive – Allocation of Math Marketing’s team within the client to understand the processes, platforms and necessary accesses.
Tagging – Validation of tagging applied in all digital environments of the financial institution. Solving problems found and suggesting improvements regarding tagging.
Data Mapping – Recognition of the processes and databases that will impact the project.
Database consolidation – Consolidation of different databases in one environment.
Data Relationship – Relationship of different databases in search of a common denominator.
Exploratory Analysis – Initial exploration of consolidated data.
KPIs – Definition of the main KPIs that will guide the model, such as Removal code number of philippines Effect, CAC (Customer Acquisition Cost), Time Decay (time to live), etc.
Check out what the journey to be analyzed was like when the project began.
Attribution Model Journey at Project Startup
Note that by analyzing the line above, and with a Last Click model in mind, we could conclude that organic access from Google is the most efficient media. Therefore, logically, I can choose to reduce some investment, or even discontinue a media/audience with a higher CAC (Customer Acquisition Cost). This is what we could call a Last Click Driven strategy, that is, targeted based on the click. Remember that this model “steals” all the merits for itself? But would a strategy guided by this model be able to answer the following question?
Do you know the impact of removing this media that you are considering turning off?
To respond safely, the following assumptions were set as guidelines for the project.
Analyze the percentage contribution of each media until conversion (in its different journeys).
Choose the media or audiences with the highest and lowest CAC and, through the KPI Removal Effect, measure the real impact of the removal on conversions (customer activation).
Optimization : Choosing to discontinue, reduce and/or reallocate investments based on Machine Learning model indicators .
With the assumptions analyzed and implemented based on our framework, the performance analysis of each media can be monitored more accurately and we can show the client the best way to reallocate their investments to ensure more conversions.
Notice in the real example below where we can see a customer's journey to conversion. He was impacted by several Google Display impressions, until he accessed the website on 09/18 . After being impacted a few more times on different websites, he ended up becoming a customer on 09/24 , right after clicking on a Google Ads keyword .
Did you notice that this example validation deconstructed the initial idea that the last click was the most important in the journey? Look at the figure again, Google Ads was the first in line in the initial journey, and in the validation, it was responsible for the conversion. This allowed us to put the attribution models to be learned through Machine Learning, since machine learning can recognize these behavior patterns easily.
The work resulted in a much more accurate identification of the most valuable media in the customer journey. And measuring it was of utmost importance for defining investment allocations. Currently, models based on artificial intelligence are learning the following principles based on our initial validated modeling.
AI-Based Attribution Models - Removal Effect and CAC
AI-Based Attribution Models - Time Decay and Journey Dynamics
Artificial Intelligence-Based Attribution Models - Descriptive Profile
Understand that even without machine learning, you can perform this analysis according to the needs of your company or your client.