Cloud-based Machine Learning Operations (MLOps) drives 2.3% revenue uplift for FinTech unicorn

Summary

A global InsurTech sought to improve device protection sales performance at cell phone retailers across Indonesia. The company connects insurers with customers through retail partners, with sales ambassadors visiting stores to train staff on insurance benefits and support the sales process.

Amdocs Cloud Studio was engaged to create a data-driven solution to improve travel planning using the cloud-based analytics platform Databricks.

• Amdocs Cloud Studio data scientists designed and implemented machine learning operations (MLOps) workflow to optimize and automate route planning
• Time sales ambassadors spend in non-productive travel was reduced by 3.5%
• Strategic optimization of store visits resulted in 2.3% revenue increase
   

Challenge

Automated route planning was a complex data challenge

The sales team's manual approach to route planning meant travel was inefficient and costly, with decisions about store visit frequency and prioritization made subjectively. A data-driven solution was needed to reduce travel time, strategically prioritize stores, and ensure optimal coverage to drive sales revenue uplift.

Addressing the challenge demanded more than a simple route optimization algorithm. With a 100-plus sales team working across the Indonesian archipelago, the solution had to account for multiple start points and destinations, traffic patterns in various towns and cities, and store opening times. Furthermore, some stores were more strategically important than others, with many requiring several visits in a given monthly schedule that needed to be spaced out effectively.

Since in-house cloud and data experts were focused on the company's core digital ecosystem, external assistance was required. Amdocs Cloud Studio was engaged to support the transition from subjective travel planning to an objective, data-driven approach. The goal was to provide automated, optimized travel schedules for each sales ambassador. Amdocs Cloud Studio designed and deployed an ML model on Databricks, the InsurTech's AWS-hosted unified analytics platform.

Solution

End-to-end ML model deployed on Databricks using MLOps

Amdocs Cloud Studio's data scientists developed an ML algorithm for deployment on Databricks using MLOps practices, covering all variables associated with the sales team's route planning.

Minimizing travel time and maximizing the time spent productively in cell phone stores involves a two-step process. Firstly, a 'store importance' value calculation prioritizes which stores a sales ambassador should visit in a given month, and how many times. Then, a 'route optimization' calculation determines the order in which they should be visited, given store opening hours, travel distance, traffic congestion, travel time (via a maps API), start and destination points, salesperson working hours, and home location.

Since the algorithm's comprehensive approach to these variables makes it compute-intensive, Amdocs Cloud Studio's data scientists needed to optimize it for efficiency and cloud cost reduction. Leveraging core aspects of Databricks' functionality in the MLOps workflow enabled scalability and further cost optimization, while governance and security were enhanced through their integration into the MLOps processes.

Finally, Amdocs established best practice MLOps methodologies for deployment on Databricks, which the InsurTech now uses for all ML use cases.