What challenges did Aviva face before AI implementation?

Deploy and operate machine learning models at scale for claims decisioning. Reduce the manual operational burden on data scientists and improve production monitoring. Standardize ML development and deployment across use cases.

What was the impact of the AI solution?

Deploying hundreds of ML use cases in weeks rather than months. Reduced infrastructure costs by 90% versus the prior on-premises ML platform. Data scientists previously spent more than 50% of their time on operational tasks.

Source

Scaled productionEvidence: Medium50/100

Aviva built a scalable MLOps platform with Amazon SageMaker for car insurance claims decisions

Use case typeClaims automationUpdated Oct 3, 2024

Aviva built a fully serverless MLOps platform to standardize machine learning development, deployment, and monitoring across insurance use cases. The first production use case supports car insurance claims decisions, combining multiple ML models, external data, and business rules to recommend repair, write-off, or further investigation. The platform was designed to reduce manual work for data scientists and create a repeatable blueprint for future ML use cases.

Organization: Aviva
Industry: Insurance
Location: United Kingdom
Published: October 2024

Reported outcomes

−90%

infrastructure cost reductionCost savings

−50%operational time spent by data scientists400,000 claims/yearinsurance claims handled annually

Strategic outcomes

Scale & capacityDeploy ML use cases in weeks rather than monthsSpeed & agilityStandardized claims ML lifecycleBetter decisions & insightAutomated claims triage recommendationsRisk & complianceImproved monitoring and auditability of model decisions

Catalog median for cost savings deployments: −40% across 177 reported metrics. Compare benchmarks →

Why do we believe this?Outcome claims, sources, and evidence checks

Normalized claim

Infrastructure cost reduction: 90% decrease

AWS Machine Learning BlogOct 3, 2024Blog postExplicit claimMedium evidence strength

resulting in a 90% reduction in infrastructure costs compared to the company0s previous on-premises ML platform solution

Normalized claim

Operational time spent by data scientists: 50% decrease

AWS Machine Learning BlogOct 3, 2024Blog postExplicit claimMedium evidence strength

data scientists spending more than 50% of their time on operational tasks

Normalized claim

Insurance claims handled annually: 400,000 claims/year increase

AWS Machine Learning BlogOct 3, 2024Blog postExplicit claimMedium evidence strength

handling approximately 400,000 insurance claims annually

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Aviva
Provider: AWS
Maturity: Scaled Production
Linked source: AWS Machine Learning Blog

Deploy and operate machine learning models at scale for claims decisioning

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 3 of 4

1Claims automation
2MLOps platform
3Workflow orchestration

Deploy and operate machine learning models at scale for claims decisioning.
Reduce the manual operational burden on data scientists and improve production monitoring.
Standardize ML development and deployment across use cases.

Implemented a fully serverless MLOps platform based on the AWS Enterprise MLOps Framework.
Used Amazon SageMaker Studio, SageMaker Projects, SageMaker Model Registry, SageMaker Automatic Model Tuning, SageMaker Experiments, and real-time inference endpoints to build a CI/CD-enabled ML lifecycle.
Orchestrated inference with Amazon API Gateway and AWS Step Functions, combining model predictions, external data, and business logic for claims outcomes.
Stored logs and artifacts in Amazon S3 and Amazon Data Firehose with AWS KMS and IAM controls.

Technologies

Amazon SageMaker AWS Enterprise MLOps Framework Amazon SageMaker Studio Amazon SageMaker Projects Amazon SageMaker model registry Amazon SageMaker Automatic Model Tuning Amazon SageMaker Experiments Amazon API Gateway AWS Step Functions Amazon SageMaker real-time inference endpoints Amazon CloudWatch Amazon Data Firehose Amazon S3 AWS KMS AWS IAM AWS CDK AWS CodeCommit

Deploying hundreds of ML use cases in weeks rather than months.
Reduced infrastructure costs by 90% versus the prior on-premises ML platform.
Data scientists previously spent more than 50% of their time on operational tasks.

Architecture

A serverless MLOps platform built on the AWS Enterprise MLOps Framework spans development, staging, and production accounts. SageMaker Studio and Projects support ML code development and CI/CD; SageMaker Model Registry governs model approval and promotion; SageMaker Automatic Model Tuning and Experiments handle hyperparameter search and evaluation; SageMaker real-time inference endpoints serve predictions; Amazon API Gateway and AWS Step Functions orchestrate claim decisions; Amazon Data Firehose and Amazon S3 feed reporting and Snowflake; AWS KMS, IAM, CodeCommit, CloudWatch, and VPC controls secure data, code, and logs.

Sources & evidence1

Evidence: Medium50/100Evidence strength