MLOps save millions in a matter of months

A US-based Fortune 100 health insurer was losing over 4,000 hours of productivity per month and facing budget overruns amounting to millions of dollars. Why? Because it was attempting to scale its operations to serve over 500 AI services, but without machine learning operations (MLOps) or an effective governance structure. This was cumbersome for IT production and operational management and wasted time and money. 

Time for transformation

The insurer discovered that Fractal’s combination of AI, engineering, and design expertise could help it define the MLOps and governance strategy it needed to operate its AI services. As a result, it set Fractal an ambitious goal: to establish the architecture and run a pilot. With massive potential savings at stake, time was of the essence.

Rising to the challenge

The Fractal team started by reviewing the insurer’s existing technology and use cases that need to be served. It identified a prioritization framework and defined a roadmap to implement a scalable MLOps framework in the business and could immediately identify the cost-saving potential. It then aligned all stakeholders, including IT operations, and specified the optimal technologies considering existing governance and IT guidelines.

The chosen technologies included an in-house automated machine learning (AutoML), a customized feature store to manage over 17,000+ features for batch and real-time services, and MLflow to orchestrate CI/CD (continuous integration and continuous deployment) as well as model registry and model versioning.

These technology components’ scalability, serviceability, and maintainability uniquely matched the client’s requirements in relation to feasibility, desirability, and viability. 

Once this had been organized, the Fractal team set up a specialist engineering unit to selectively scale high-priority AI services in an MLOps framework. It then deployed and tested the pilot service end-to-end and ramped up the MLOps services to incorporate other scalable technologies.

Rapid results

A design-driven architecture and strategy enabled the Fractal team to deliver the engagement rapidly. With a further investment of 24 weeks, the pilot engagement on MLOps was complete, and the client organization realized significant time and cost savings. The ongoing scaling of further ML models was reduced from months to weeks.

The future is now bright. The Fractal implementation has opened a path to fast and cost-effective expansion of the insurer’s AI services, enabling it to meet the needs of over 40 million customers in real-time and at scale.