Edelweiss improves cross-sell using machine learning on Amazon SageMaker

Edelweiss Tokio Life Insurance Company Ltd, a life insurance company in India, built a data-driven cross-sell solution to replace rule-based recommendations that were producing irrelevant offers and missing conversion opportunities. The implementation used Amazon SageMaker to train and tune a CatBoost-based cross-sell propensity model, used a separate SVD-based policy recommendation model, and added frequent pattern mining to validate popular policy bundles. The team processed about 100,000 records with more than 200 attributes, standardized the training and batch inference workflow with SageMaker managed training, automatic model tuning, ECR-hosted custom containers, notebooks, and batch transform, and deployed the model in production for batch scoring.

Industry
Insurance
Location
India
Published
June 2021

Reported outcomes

+200%

cross-sell rate increaseRevenue & growth

+75%conversion from high-propensity segment+80%model recall+40%model precision+88%top-decile cumulative cross-sell coverage

Strategic outcomes

Customer experience & trustReduced irrelevant cross-sell offersNew product / capabilityBuilt machine-learning cross-sell recommendationsNew product / capabilityCreated policy ranking recommendation modelSpeed & agilityStandardized model development workflow

Catalog median for revenue & growth deployments: +34% across 150 reported metrics. Compare benchmarks →

Primary read

Use case focus

Showing 3 of 3

  • 1Cross-sell optimization
  • 2Recommendation system
  • 3Predictive modeling
  • Improve cross-sell recommendations in life insurance.
  • Reduce irrelevant or unsolicited offers from rule-based cross-sell.
  • Adapt recommendations to changing customer behavior and product features.
  • Built a supervised cross-sell propensity model in Amazon SageMaker using CatBoost.
  • Built a separate policy recommendation model using SVD to rank the top three policies for each prospect.
  • Used frequent pattern extraction with Apriori and FP-Growth to identify common policy bundles and validate recommendations.
  • Standardized model development and inference with SageMaker managed training, automatic model tuning, ECR-hosted custom training containers, notebooks, and batch transform.
  • Cross-sell rate increased by 200% compared with the previous financial year.
  • Achieved 75% conversion from the high-propensity segment.
  • Recall exceeded 80% and precision exceeded 40% after feature engineering and tuning.
  • The first three deciles captured 88% of cross-sell cases, improving prioritization of prospects.
  • SageMaker standardized and accelerated model development, inference, and maintenance workstreams.
Architecture

Two-pipeline ML architecture: a cross-sell propensity classification model built with CatBoost on Amazon SageMaker for decile ranking and batch inference, plus a separate SVD-based recommendation model for top-policy ranking. The solution also used ECR-hosted custom training containers, SageMaker automatic model tuning, batch transform, and frequent pattern mining (Apriori and FP-Growth) for validation of recommendations.

Implementation partners1
Sources & evidence1
Groundedness: 5/5Type: Blog PostPublished: Jun 2, 2021Publisher: AWS Machine Learning BlogEvidence: VendorConfidence: Medium

AI-generated summary. Verify important details with the linked sources before relying on this case.

Explore related AI use cases

Was this useful?