Kyndryl Teradata Gen AI for Insurance Claims Automation

Kyndryl + Teradata’s First Notice of Loss (FNOL) solution modernizes insurance claims intake and adjudication on AWS. It unifies omni-channel claim initiation, intelligent document processing, GenAI summarization, and fraud detection—reducing cycle time while improving accuracy and auditability.

How it works on AWS

Customers initiate claims via a web portal or Amazon Connect IVR. Evidence (photos, forms, police reports) is uploaded through pre-signed Amazon S3 URLs. Amazon GuardDuty scans payloads on arrival, and AWS Lambda orchestrates the downstream pipeline. Documents are automatically processed with Amazon Textract (OCR), Amazon Comprehend (PII, entities, key phrases), and Amazon Rekognition (image classification and object detection). Amazon Bedrock (e.g., Titan FM) generates an adjuster-ready narrative and document checklist status, while Amazon SageMaker supports fraud-risk scoring. Claim metadata lands in Amazon Aurora PostgreSQL with encryption and secrets managed by AWS KMS and AWS Secrets Manager. Integration patterns use Amazon API Gateway and (optionally) Amazon EventBridge for event flow.

Intelligent analytics with Teradata

Teradata VantageCloud Lake augments the AWS-native flow with vector search/RAG and cross-claim analytics. An enterprise vector store indexes embeddings from claims documents to speed retrieval, similarity search, and investigation. ClearScape Analytics (BYO-LLM) supports explainable reasoning and federated queries across structured and unstructured data.

Key AWS services

• FNOL intake: Amazon Connect; web on Amplify/EC2
• Upload & security: Amazon S3 (pre-signed URLs), Amazon GuardDuty
• Orchestration: AWS Lambda (and AWS Step Functions where required)
• AI/ML & NLP: Amazon Textract, Amazon Comprehend, Amazon Rekognition, Amazon Bedrock, Amazon SageMaker
• Data & security: Amazon Aurora PostgreSQL, AWS KMS, AWS Secrets Manager
• Integration & observability: Amazon API Gateway, Amazon EventBridge, AWS CloudTrail, Amazon CloudWatch

These service mappings align to intake, processing, ML, data management, and analytics layers.

What buyers get

• Faster time-to-decision: Automated document checks and GenAI summaries reduce manual handling for adjusters. • Improved fraud detection: Amazon SageMaker models score risk; vector search surfaces similar historical claims for investigation. • Security & compliance by design: PII detection/redaction, encryption at rest, secrets management, and full audit trails via CloudTrail/CloudWatch. • Scalable, serverless core: AWS-managed services minimize ops overhead and scale with volume. • Extensibility: Modular APIs and Teradata’s query fabric enable additional use cases (e.g., subrogation, SIU, cross-policy insights) without re-platforming.

Typical flow

1. Customer submits FNOL via web/IVR
2. S3 upload (pre-signed URL) with GuardDuty validation
3. Lambda triggers Textract/Comprehend/Rekognition
4. Bedrock creates claim summary; SageMaker scores fraud risk
5. Aurora stores results; human-in-the-loop review as needed
6. Teradata builds embeddings for retrieval and analytics.

Deployment & fit

Designed for carriers, MGAs, and brokers seeking to speed claims intake, cut cost, and enhance fraud detection with AWS-native foundations plus enterprise-grade analytics from Teradata. The solution emphasizes rapid build and minimal integration effort, and can be extended for broader intelligent document processing in FSI.