Paid Media - From last click to multi-touch attribution

The problem with paid media

Our client, a Brazillian Fintech, depended on a significant investment in paid media and the purchase of audiences for its growth strategy and measured everything in terms of first click and last click. There was no complete view of the shopping journey; in the words of the Client himself: “the campaign tests are done a little blindly…”
When they hired Math, the client expected to have a model that would bring transparency, as well as the measurement of real investment results and insights of what would be the possible ways to improve the outcome.

The multi-touch attribution model for different media sources

The goal of the model is to have a real view of the entire customer journey, considering paid media (extremely diverse set with Google Adwords, Facebook Ads and National TV News), organic traffic (leveraged by a consolidated content marketing strategy) and direct traffic with. In addition, we were to build a vision of conversion per audience to justify, optimize, reduce, or change the direction of investments of performance team.

Solutions to unify the Customer Journey

First, we delved into all the data available to understand the structure and content of what was being stored. We identified all the data capture gaps of the journey, creating an ideal model to be used by media, advertising and performance areas with tagging and parameterization criteria.
After centralizing the data in a single point, in this case, a Data Lake, we correlated the data and evaluated which were the most relevant for the construction of a multiple impact attribution model.
We built the Machine Learning model, revealing the relevance of each of the media (even those that did not generate clicks), the cost of acquiring the complete workweek executed by customers and the possible effect of the relocation of investments or also removing any of channels.

Want to know the results?

Want to know how we do it?


Methodology for a multi-touch attribution model

Understanding the processes, platforms and access needed, such as:
• Validation of tagging;
• Error adjustments;
• Improvement suggestions;
• Understanding methods and databases that will impact the project;
• Consolidation of different databases in one environment;
• Relationship of different bases in search of a common denominator.
For the direct non-viable relationships (for example, with user ID), a propensity model was created that relates them by inference, ensuring the use of the most significant volume of relevant data possible. This makes it possible to monitor the following indicators:

• Removal Effect;
• CAC - Cost of acquring a Customer
• Time Decay for each impact and average per channel
• Journey Dynamics

want to boost your business?

send us a message and we will get back to you as soon as possible!