Enhance customer lifetime value for a mobile app by leveraging bayesian inference
Our client, a mobile app company, sought to optimize their customer engagement strategies to maximize customer Life Time Value (LTV). Understanding the factors that influence customer LTV enable businesses to forecast future revenue from customers but most importanlty help them to devise strategies to enhance their customer value.
Leveraging bayesian inference, we quantified the impact of various factors on LTV, enabling the client to make data-driven decisions and implement effective strategies.
Approach
Data Collection and preparation
We began by collecting historical data on customer acquisition costs, engagement metrics, and observed revenue. Some key features of the dataset included:
channel_acquisition
: The acquisition channel used, such as TikTok, Instagram, YouTube, TV and searchcampaign_name
: the utm of the campign along with key descriptions such as creative type and support type.time_{X}
: Various time-related features, including time of acquisition, time of download, and time of subscription.acquisition_cost
: The amount spent to acquire each customer.engagement_score
: A composite metric representing user interactions, session frequency, and feature usage within the app.revenue_X_month
: The total revenue generated by each customer over specific periods (e.g., revenue_1_month, revenue_3_month, etc.).
These detailed features allowed us to build a comprehensive predictive model and gain a nuanced understanding of the factors driving Customer Lifetime Value (LTV).
Model specification
Using PyMC, a Python library for probabilistic programming, we specified a bayesian regression model where LTV is a function of our different variables.
One might wonder why we opted for a linear formulation rather than employing more powerful methods such as Gradient Boosted Trees (GBT). Our stakeholder wanted clear and actionable insights, primary objective was to map out the relationship between actionable variables that significantly impact revenue in a manner that is clear and understandable for the business team. A linear formulation offer the following:
Interpretability: Linear models provide straightforward and easily interpretable results. Each coefficient in the model directly represents the expected change in the dependent variable (LTV) for a one-unit change in the predictor variable, holding all other variables constant. This clarity is crucial for business teams who need to understand and act upon the insights.
Transparency: The simplicity of linear models makes them more transparent. Business stakeholders can see how each feature contributes to the final prediction, fostering trust in the model’s outputs and facilitating easier communication across teams.
Actionability: By using a linear model, we can clearly identify which variables have the most significant impact on LTV and quantify this impact. This direct mapping helps the business team to prioritize actions and strategies effectively.
Trade-offs
While more complex models like GBT can capture non-linear relationships and interactions between variables, they often do so at the expense of interpretability. These models can act as “black boxes,” making it challenging to explain the rationale behind predictions to non-technical stakeholders.
Conclusion
Our client gained a nuanced understanding of how acquisition and engagement efforts impact Customer Lifetime Value. Our method provided actionable insights enabling the client to optimize their strategies and achieve significant improvements in LTV.
This case study exemplifies the power of advanced data science techniques in transforming business strategies and achieving tangible outcomes. Predictive Lifetime Value (LTV) models are especially valuable for understanding how actions taken by teams, such as acquiring and engaging customers, influence important outcomes like revenue. These models help map the relationship between what teams do and the value they generate.