Zayo is a telecommunications company providing the fiber to connect companies and data centers over the internet, as well as the hardware and software infrastructure at both ends of that fiber. Established in 2007, Zayo has grown aggressively through strategic acquisition – 47 since its inception, including five in the past three years. These acquisitions have culminated in over 47,000 miles of fiber, connecting thousands of on-net buildings and over 100,000 devices under management throughout the U.S. and Canada.

The challenge

To expand the overall reach of Zayo’s total offering and further differentiate its value proposition in the marketplace, it acquired three unique managed service providers (MSPs) between 2021 and 2023. Each Edge MSP business unit (BU) entity brought distinct day-to-day processes, service offerings, and data infrastructures. Although the Zayo data science teams had previously developed a valuable predictive churn model for the company’s traditional fiber services, there remained a need to predict churn for their new MSP BU.

Considering the current state of the collective data and the inherent differences among the MSP entities, the Edge MSP BU lacked the ability to clearly understand what retention and churn metrics looked like in the aggregate. Consequently, it couldn’t make informed decisions or proactively take action to understand retention and mitigate churn. Also, the variance in data infrastructure, domain owners, and general way of doing business resulted in an environment that required detailed coordination, planning, and execution across business and technical stakeholders to deliver a consolidated churn model tailored for the MSP organization.

The solution

DAS42 understands all problems are business problems first and technical problems second. To develop a robust churn prediction model, we had to start by engaging with the people with the most significant insight into the key performance indicators (KPIs), metrics, and indicators that predicted churn in managed services. We conducted in-depth joint discovery sessions with sales professionals, customer success managers, technical success managers, and other subject matter experts from each of the three distinct MSPs. The result was a comprehensive feature list with the highest likelihood of producing a robust predictive model and fostering organizational buy-in.

Teams from each MSP then curated the data for each feature and uploaded their datasets to Azure Blob Storage. With a common data structure in place, DAS42 then set up data pipelines, orchestrated by Airflow, to combine the data into a common view in Snowflake. We made this view available to the Zayo Data Science team. Leveraging the features and functionality available in Dataiku, the Data Science team was able to quickly generate multiple models trained on the consolidated data available in Snowflake. Through principal component analysis, they pinpointed features with the highest predictive power and collaborated with the DAS42 team to refine the pipelines further.

The pipeline automatically ingested data from the different MSPs into Snowflake, merged the multiple datasets, and presented the consolidated dataset to the Dataiku model for monthly churn predictions after the model went through training and productionization. While the solution’s efficacy is still under evaluation, the model itself demonstrated very high (>90%) accuracy and precision. The framework implemented by DAS42 also addresses the challenges of the solution being capable of scaling with the addition of new acquisitions and supporting the iterative and evolving nature of data science.

How did it turn out?

Zayo now possesses the capability to generate predictive churn insights for its MSP organization. By leveraging a model designed specifically for the MSP business, Zayo can achieve higher predictive accuracy, resulting in more actionable insights.

This MSP churn model provides for:

• Deep predictive insights into expected top-level revenue and bottom-line profitability as impacted by projected churn
• Highlights at-risk accounts, enabling the customer success and sales teams to take proactive remediation actions to prevent or reduce the probability of churn

With the ultimate goal of retaining current customers, this model can:

• Increase customer lifetime value (LTV)
• Decrease customer acquisition cost (CAC)
• Maintain revenue streams
• Improve customer satisfaction (CSAT) and net promoter scores (NPS) with proactive customer care