AI Data Driven Attribution
Last updated
Last updated
Attribution analysis is a cornerstone of modern digital marketing, allowing businesses to track and evaluate the impact of various marketing channels on conversions. As customers interact with a brand across multiple touchpoints, such as social media ads, search engine results, and affiliate programs, understanding which channels contribute most to driving conversions is essential. Attribution analysis provides this understanding to help marketers optimize their strategies and budgets.
Below, we provide a basic definition of attribution analysis, explain its importance in modern marketing, and explore the different attribution models you may encounter, from simple heuristic models to more complex data-driven approaches. Later, we explain the model used at Roivenue, highlighting how it processes data and assigns credit to various marketing touchpoints.
Attribution is a method used to determine the contribution of different marketing channels to a conversion in a customer journey. Attribution is crediting channels with value according to the rules of the used Attribution Model.
Marketing channels (such as social media, paid search, affiliate programs, and more) are represented by touchpoints (impressions and visits). These touchpoints in the customer journey are moments of interaction between the brand and the customer. In the picture below, touchpoints are displayed by colorful dots. Conversion customer journey happens when the potential customer converts into a customer by making a purchase or performing the desired action, such as signing up for a demo or an email list. Once the conversion occurs credit is assigned to the different touchpoints based on the rules used within the chosen attribution model.
Attribution models can be categorized into two types: heuristic and data-driven.
Heuristic models are based on predefined rules that determine how credit is assigned to different touchpoints within the customer journey. These models are simple and do not consider actual data, making them less accurate but easy to implement.
Types of Heuristic Models:
First-Touch Attribution
This model assigns 100% of the credit to the customer's first interaction with the brand. It is useful for understanding which channels are effective in creating initial awareness.
Last-Touch Attribution
This model gives all the credit to the last interaction before the conversion. It helps identify which touchpoints are most effective in closing sales or generating leads.
Linear Attribution
This model evenly distributes credit across all touchpoints in the customer journey. It provides a balanced view but may not accurately reflect the varying impact of each touchpoint.
Time-Decay Attribution
Credit is assigned based on the recency of the touchpoints, with more recent interactions receiving more weight. This model recognizes that the closer a touchpoint is to the conversion, the more influence it likely has.
In today's rapidly evolving technological landscape, data-driven models have become indispensable tools across industries. These models, built on vast amounts of data, enable organizations to make informed decisions, predict future trends, and optimize operations.
At their core, data-driven models are algorithms and mathematical frameworks that leverage historical data to predict outcomes, identify patterns, and guide decision-making processes. Unlike traditional models that may rely heavily on theoretical assumptions or expert judgment, data-driven models are grounded in real-world data. This allows them to adapt to new information and improve their accuracy. Crediting of the channels using data-driven models is based on the channels' ability to influence potential customers to convert.
The data-driven model includes, for example, the Shapley Value, the Markov model, or a group of AI-based models. In Roivenue, we have an AI-based attribution model based on a recurrent neural net (RNN).
Recurrent Neural Network (RNN) Model
At Roivenue, we use the more advanced Recurrent Neural Network (RNN) model, which we now refer to as our AI Data Driven model. While the earlier statistical models were a significant step forward, their limitations became apparent over time. For instance, these models required a large number of conversions from each channel to be accurate, making them less practical for smaller e-commerce projects and detailed campaign-level attribution.
Furthermore, the one-size-fits-all nature of Markov and Shapleyβs models lacked the flexibility to adapt to the unique needs of different clients, sometimes allowing less valuable channels to be overrepresented in the attribution mix. The AI Data-Driven model addresses these challenges by offering more flexibility and requiring less data. This allows for more precise and timely insights, which are crucial for optimizing marketing strategies daily. The model is also highly customizable, ensuring that it better reflects the specific dynamics of each clientβs marketing channels and campaigns.
In the following text, we will examine this model and its applications in attribution analysis.
The model based on RNN is designed to learn from historical data and predict future outcomes based on the identified patterns. It is trained on vast customer journey data, allowing it to understand and recognize the sequences that lead to conversions. It calculates the probability of conversion for each touchpoint in the customer journey and allocates the credit for each touchpoint. However, before the model can perform these tasks, it must undergo several critical steps, including data collection and model training. Below, we explore each of these steps in detail.
The first step in our AI Data Driven Model process is collecting the necessary data. We begin by extracting traffic data from your web using Roivenue Measurement. This data is then enriched with information from advertising platforms such as Google Ads, Meta Platforms, Adform, Bing, and others.
In addition to integrating data from websites and advertising platforms, it is also possible to connect data from an Order Management System (OMS) to Roivenue. This data is referred to in Roivenue as the Transaction Feed. Once connected, the resulting metrics, such as net revenue, are adjusted to account for returns and order cancellations, providing a more accurate and comprehensive view of your performance.
With this enriched data, we construct customer journeys. Each touchpoint has its parameters (characteristics) in the customer journey. Some standard parameters are the time of the touchpoint (timestamp), source, medium, number of page views, touchpoint type, conversions (= credit assigned to the touchpoint), and more. The model can also consider additional parameters if necessary.
If your Roivenue was implemented before July 2024, it is possible you are still running a version based on Google Analytics 4 data instead of Roivenue Measurement. That limits how the customer journey is reconstructed but appart from that, all the other steps are the same.
With the collected data ready, the next step is to train the RNN model. During training, the model learns from historical customer journeys to identify patterns that are increasing the likelihood of a conversion.
To ensure the predictions remain accurate and reflect the latest data trends, the model is retrained every 14 days. This frequent retraining guarantees that the model stays up-to-date with any changes in customer behavior, thereby ensuring the relevance and accuracy of the predicted data.
This training phase is iterative, with the model continually refining its understanding as it processes more up-to-date data, ensuring that it remains aligned with the current market dynamics and customer interactions.
After training the model, it estimates the conversion probability at each touchpoint within a customer journey. For every customer interaction with your brand, the model evaluates how likely that interaction lead to a conversion.
In addition to the sequence of touchpoints and the time between the touchpoints, the model also considers specific parameters (or characteristics) of each touchpoint, such as the type of interaction, channel, or engagement level. These parameters influence how the model perceives the significance of each interaction with the others in that specific customer journey.
The probability estimation is based on patterns the RNN model has learned during training, where it processes sequences of touchpoints and identifies how each interaction in the journey increases the probability of a conversion. These probabilities help to understand how each touchpoint contributed to the final conversion.
The final step in the process is credit allocation. Once the model has estimated the conversion probability for all touchpoints, it distributes the credit for the conversion across these touchpoints according to those probabilities. Each touchpoint receives a share of the credit based on how much it contributed to the final conversion. The sum of all credits within a single customer journey always equals 1.
