Table of Contents
Improving customer retention is a common goal for businesses, and data scientists often receive requests to help with this task. However, traditional machine learning approaches that focus on predicting churn often fall short in making a significant impact on retention. These approaches lack the ability to recommend optimal next steps or interventions for improving customer retention. Additionally, they lack explainability and trust, particularly among marketing executives.
Challenges with Traditional Approaches
Traditional machine learning approaches for customer retention typically involve building predictive models based on observed data about the customer base. Once a prediction is made, post-hoc explainability methods are used to understand the importance of each feature within each prediction. However, these approaches often fail to consider the causal relationships between variables and rely solely on statistical associations.
One of the fundamental challenges of using statistical associations is that it is difficult to determine if these associations are causal or simply spurious. For example, a correlation may be found between offering discounts and churn rate. However, a causal analysis may reveal that dissatisfied customers are more likely to complain, leading to them being offered a discount. This confounded relationship is not captured by a correlational approach.
Furthermore, statistical associations do not provide any information about the effects of actions. They only describe the relationship between variables without understanding how they affect each other. For example, a feature importance ranking may suggest that offering discounts should no longer be used as a churn reduction mechanism. However, this approach fails to reason correctly about the best actions to be taken to reduce churn.
Introducing Causal AI
To overcome the limitations of traditional approaches, a causal approach can be used to create decision workflows for marketing and sales teams. Causal AI goes beyond simple predictive models and incorporates causal graphs and structural causal models (SCMs) to provide explainable and actionable insights for improving customer retention.
Building Decision Workflows with Causal AI
Building a causal customer retention decision workflow involves several steps:
-
Create a Causal Graph: Using Human-Guided Causal Discovery, a causal graph is created to uncover cause and effect relationships within the dataset.
-
Build a Structural Causal Model (SCM): An SCM is created based on the causal graph to predict which clients are likely to churn and why. Unlike traditional models that only map relationships between inputs and outputs, an SCM models relationships between all variables, including indirect and second-level relationships.
-
Use Causal Decision Intelligence: Causal decision intelligence is applied to the SCM to understand the effects of interventions and provide optimized recommendations for specific actions based on business constraints.
-
Create and Deploy a DecisionApp: A decisionApp is created using a framework like Dara, which allows business stakeholders to interact with the modeling work and take action based on the recommendations.
The Causal Data Science Workflow
The causal data science workflow consists of several key steps:
-
Causal Discovery: Prior to building a predictive model, causal discovery is carried out to uncover cause and effect relationships. This process combines algorithmic approaches with domain knowledge to build a causal graph that describes relationships between variables.
-
Algorithmic Causal Discovery: Algorithmic causal discovery involves applying causal discovery algorithms to the data. decisionOS, a platform for Causal AI, provides over 15 causal discovery algorithms that can be applied selectively depending on the structure of the data and expert knowledge.
-
Domain Expert Review: The causal graph generated through algorithmic causal discovery is then presented to a domain expert for review. The expert can provide feedback and suggest changes to the graph using tools like the Human-Guided Causal Discovery App.
-
Creating the Structural Causal Model: Once the causal graph is finalized, the data scientist creates the structural causal model. This model uses the causal parents of each feature in the graph to build a predictive model of the entire system. This allows for a better understanding of the relationships between variables and their impact on the target variable.
Benefits of Structural Causal Models (SCMs)
Structural causal models offer several benefits compared to traditional models:
- Predictions are more robust and fully explainable, as the model incorporates causal relationships between variables.
- Interventions and counterfactuals can be computed, allowing for a better understanding of how changes to a process can affect the outcome.
- Causal decision intelligence engines can be applied to the predictive model, enabling the selection of optimal interventions based on business constraints.
Decision Intelligence Engines
Decision intelligence engines enhance the capabilities of SCMs by generating actionable business recommendations. In the context of customer retention, algorithmic recourse is particularly useful. Algorithmic recourse allows users to determine the most cost-effective intervention, given a set of possible actions and business constraints, to prevent churn.
App Building with decisionOS
decisionOS provides a framework for building decisionApps, which can be used by business stakeholders to interact with the modeling work and take action based on the recommendations. The decisionApp framework, Dara, is a Python-based framework optimized for Causal AI. It includes high-level components that can be easily integrated into the decisionApp, and custom pages can be created to tailor the application to the needs of the business user.
Conclusion
Using Causal AI in marketing and sales teams can significantly improve customer retention by providing optimal interventions and actionable recommendations. By incorporating causal graphs and structural causal models, data scientists can create decision workflows that are explainable, robust, and effective. decisionOS, a platform for Causal AI, offers a comprehensive set of tools and capabilities to support the entire causal data science workflow, from causal discovery to building SCMs and deploying decisionApps. With Causal AI, businesses can make informed decisions and take actions to maximize customer retention and drive growth.
