ARC SENTINEL

Inspiration

This project is inspired by Arca Continental's value of "Vocation of Service."

We realized that churn is not just a data problem—it is a business problem that directly affects customers, sales teams, and revenue. While many organizations react after a customer has already stopped buying, we wanted to build a solution that empowers commercial teams to act earlier, make informed decisions, and better serve their customers.

Our goal was to transform historical data into actionable intelligence that helps prevent customer loss before it happens.

Besides, we wanted to focus on developing analytical skills to contribute on our technical skills development.

What it does

ARC Sentinel is an AI-powered Commercial Intelligence Control Tower designed to help Arca Continental proactively manage customer churn.

The platform:

Predicts customer churn risk using machine learning. Detects early warning signals before abandonment occurs. Explains why a customer is at risk through Explainable AI. Identifies risk concentration across territories and customer segments. Estimates revenue at risk. Prioritizes customers based on business impact. Recommends commercial actions to improve retention.

Rather than simply answering "Who is likely to churn?", ARC Sentinel helps answer:

Why is this customer at risk? How urgent is the situation? What action should be taken first? Where should commercial resources be focused?

How we built it

We built ARC Sentinel using a combination of machine learning, business intelligence, and modern web technologies.

Our solution includes:

LightGBM for churn prediction scoring. SHAP (SHapley Additive Explanations) for model interpretability. Feature engineering based on customer behavior, sales trends, purchase frequency, refrigeration equipment, and operational indicators. Context-aware analysis incorporating environmental and market factors such as seasonality, weather conditions, and local business activity. A modern dashboard experience inspired by enterprise analytics platforms.

The platform architecture follows a pipeline:

Data → Churn Prediction → Explainability → Prioritization → Recommended Actions → AI Commercial Assistant

Challenges we ran into

One of our biggest challenges was realizing that churn is not always caused by customer dissatisfaction or abandonment intentions.

During our research, we discovered that external factors such as weather, seasonality, economic activity, and local conditions can significantly influence purchasing behavior.

This forced us to rethink the problem. Instead of building a simple churn predictor, we focused on distinguishing between temporary fluctuations and genuine churn risk.

Another challenge was balancing predictive performance with explainability. Commercial teams need to trust recommendations, so transparency became just as important as model accuracy.

Accomplishments that we're proud of

Reframed churn prediction as a broader commercial intelligence problem. Built an end-to-end prototype that goes beyond scoring and provides actionable recommendations. Incorporated Explainable AI to make predictions understandable and trustworthy. Designed a business-oriented experience focused on decision-making rather than raw analytics. Validated our approach through feedback from an experienced data analyst working with predictive analytics and business intelligence solutions.

What we learned

This project taught us that successful AI solutions are not only about building accurate models.

We learned the importance of:

Understanding business context. Designing for decision-makers rather than data scientists. Combining predictive analytics with explainability. Translating technical outputs into actionable business recommendations.

Most importantly, we learned that preventing churn requires understanding why customers change their behavior, not just predicting that they will.

What's next for ARC SENTINEL

Our next steps include:

Integrating real-time commercial and operational data streams. Expanding contextual intelligence with weather, economic, and competitor signals. Implementing anomaly detection for emerging market risks. Developing personalized retention strategies using generative AI. Adding automated workflow integrations for commercial teams. Deploying a scalable version capable of supporting multiple regions and business units.

Our long-term vision is to transform ARC Sentinel into a comprehensive commercial intelligence platform that helps organizations move from reactive decision-making to proactive revenue protection.