MLOps Framework Solution to ML Business Problem

Prior Situation / Scenario:

  • Business Problem: increasing number of subscribers with involuntary churn.​
  • Over 50% of machine learning projects are never put into production due to labor intensive and inconsistent workflows.
  • Manual ML model improvement in response to performance deterioration.
  • Lack of scalability and high impact of staff turnover due to key-person dependency.

Client Challenges:

  • Estimate daily propensity to involuntary churn (score/prediction) for each customer to improve Retention & Collection campaigns.
  • Manage the lifecycle of the ML solution.
  • Standardize operationalization.
  • Enable or disable functionality remotely without deploying code​.
  • Automatic and self-healing ML model (retraining).
  • Misuse of talent: high amount of manual task and low business impactful tasks.

Strata Solution/ Key Enablers:

  • High-performance leading-edge ML algorithm portfolio
  • ML Solution as a Product
    • Deployed in customer cloud account​
    • ML Model Development and Operationalization Manual ​
  • Serverless and Event Driven System​
  • Agile development: MVP + incremental functionalities

Outcome:

MLOps Framework:

  • High scalability and high availability​ architecture
  • Delivered as a product with capabilities to: Data Process, Prediction, Monitoring, Automatic Retraining (model optimization).
  • Time to market: one click deployment into production.
  • Agile improvements: granular releases into production (MVP)
  • Standardized, easy control, operation and maintenance​ of ML Solution​

Results:

  • 10% increase in Retention & Collection campaigns.
  • 90% reduction in operational manual tasks and unintentional errors.
  • 70% less effort in later releases​ (new markets) due to replicable and scalable architecture.
  • 100% focus of the analytical team on model performance development and improvement (effective use of talent).
  • 95% reduction in time to deployment.

Posted

in

by

Tags: