Back to all cases Development of an advanced 'Best-Next-Action' orchestrator for T-Mobile
Many customers become frustrated when companies fail to communicate effectively or at the right time, which can lead to contract cancellations. To prevent this and improve customer engagement, innovative companies are deploying machine learning for a Next Best-Action-platform and T-Mobile has taken precisely this step.
The challenge
Preventing poor customer communication and minimizing customer churn
Switching to real time machine learning to process customer data instantly and continuously feed ML models
Ensuring continuity: developing and building BNA Orchestrator v2 while v1 remained operational, combined with the challenges of working from home
The solution
Microservices architecture on AWS: Built using Python, FastAPI, and Uvicorn on an ECS Fargate cluster, with Docker images in ECR and CI/CD via GitLab.
Infrastructure-as-Code & Data Management: Deployment using Terraform; Pydantic for data models; JSON for data exchange; Parquet for storage; and AWS Kinesis Firehose with Glue schemas for ML model retraining.
Robust security: Well-defined IAM roles, API Gateway with WAF, and a secured environment without direct internet access, utilizing VPC endpoints and a Transit Gateway to T-Mobile’s internal network.
The result
- Revenue growth
- Real-time processing
- MLOps maturity
- Improved customer engagement