Uderstanding and predicting revenue streams is crucial for sustainable growth. For a mobile-app with a monthly subscription model, the ability to forecast cash flow accurately can significantly enhance strategic decision-making. This article delves into how we leveraged data science to create a predictive cash flow model based on Customer Lifetime Value (CLV).

Business Context

Our client, which offers a premium subscription service, faced challenges in accurately forecasting future revenue. With a steady influx of new subscribers and a monthly churn rate, predicting cash flow was complex. Traditional financial models fell short in capturing the nuances of customer behavior and market dynamics. To address this, our cleint turned to data science, aiming to build a robust model that incorporates customer acquisition costs, churn rates, subscription fees, and market trends.

Methodology

1. Data collection and preparation: We began by collecting historical data, including customer signups,in-app activity, churn rates, monthly revenue, and customer acquisition costs (CAC). This data served as the foundation for our predictive models.

2. Churn prediction: To understand and predict churn, we employed a GradientBoostedTree classifier, a machine learning algorithm known for its accuracy in classification tasks. By training the model on historical data, we were able to predict the probability of churn for each customer for the next month. This step was crucial as churn directly impacts the customer lifetime value and, consequently, revenue forecasts.(see article on ltv value where a better version of this model has been implemented).

3. Cohort analysis: We conducted cohort analysis to observe the retention patterns of customers based on their signup dates, in-app activity and subscribes service. This analysis provided insights into how different customer groups behaved over time, allowing us to refine and segment our churn predictions further.

4. Time series forecasting: For predicting future customer growth, we utilized the Exponential Smoothing method. This method helped us forecast the number of new signups each month, accounting for seasonality trends business context..

5. Cash flow modeling: Integrating all these elements, we built a comprehensive cash flow model. The model projected the number of customers, revenue, CAC, and net cash flow for next 3 year. It adjusted for inflation rates and market growth, providing a realistic forecast of future financial performance.

Challenges

Data quality and feature-store: Ensuring high-quality, consistent data was a significant challenge. We had to merge marketing database and the web-app backend database to find relevant in app activity in order to generate relevant feature.

Model complexity: Balancing model complexity and interpretability was crucial. As usual complex models might offer higher accuracy but they can be difficult to interpret and communicate to stakeholders.So we tend to not overcomplicate our model.

Dynamic market conditions: The mobile-app market is highly dynamic, with fluctuating growth rates and customer behavior.Our model needed to be adaptable to these changes to remain relevant.

Success and outcomes

Improved Accuracy: Our predictive model significantly improved the accuracy of revenue forecasts. By incorporating machine learning, we were able to capture the complexities of customer behavior more effectively than traditional methods.

Enhanced decision-making With a clearer understanding of future cash flows, the finance team could make more informed decisions regarding budgeting, marketing spend, and resource allocation. This led to optimized investments and better financial planning.

Stakeholder confidence: The data-driven approach boosted stakeholder confidence in our financial projections. Transparent and well-founded forecasts fostered trust and facilitated more productive discussions with investors and board members.

🚀 If you’re interested in learning more about how data science can transform your business forecasting or want to explore similar projects, feel free to reach out.