A specialty property and casualty carrier operating in the non-standard auto and cross-border vehicle insurance market had a claims intake problem that most carriers are still living with: first notice of loss was being collected through a static web form and routed to a human inbox for manual triage. Every claim — regardless of complexity, fraud risk, or urgency — started the same way, with someone reading a form and deciding what to do next.
The carrier's policy management platform, which handled rating, billing, and policy administration for thousands of policyholders across multiple states, had no automation layer connecting it to claims intake. The result was a slow, error-prone, and unscalable process that grew linearly with claim volume — and offered no visibility, no fraud detection, and no compliance infrastructure as AI regulation began to accelerate across US insurance markets.
QyrosCloud designed and implemented a generative AI-powered omnichannel claims processing platform on AWS — replacing the manual FNOL process with an intelligent, multi-channel intake system capable of receiving claims via web application, chatbot, SMS, and voice, then automatically triaging, analyzing, and routing each claim through a structured adjudication workflow. The result is a platform that handles the full claims lifecycle from first notice to settlement notification — with fraud detection, document intelligence, AI-assisted adjudication, and compliance-ready audit trails built in from day one.
What was at stake.
The carrier's existing claims process reflected an industry-wide reality: most carriers in the non-standard auto segment still operate FNOL through forms, email, and phone calls — not because they don't recognize the problem, but because modernizing claims infrastructure has historically required either a full platform replacement or a multi-year integration project. Neither was viable for a mid-market carrier operating on a legacy policy admin system.
“A specialty carrier serving high-risk and cross-border auto markets was processing first notice of loss through a public web form and an email inbox. There was no automation, no fraud scoring, no audit trail — just manual triage on every claim. We replaced the entire intake and adjudication workflow with a generative AI-powered, omnichannel claims platform on AWS. The inbox is gone.”
How we solved it.
QyrosCloud implemented an omnichannel claims processing solution powered by Generative AI — a production-grade architecture built on AWS-native services — and extended it with carrier-specific integrations, compliance infrastructure, and connection to the carrier's existing policy administration system. The platform was deployed using AWS CDK for repeatable, auditable infrastructure-as-code deployment.
The first and most visible transformation was replacing the single static web form with a true omnichannel intake capability. Policyholders can now initiate a claim through whichever channel they prefer — and all channels feed a unified, structured claims record.
- Web application: A React-based claims portal deployed via Amazon CloudFront, protected by Amazon Cognito authentication and AWS WAF, providing policyholders with a guided digital FNOL experience
- AI-powered chatbot: An Amazon Lex conversational bot embedded directly in the web portal, enabling natural language claims initiation with structured data capture
- SMS: Two-way SMS via Amazon Connect and AWS End User Messaging, allowing policyholders to initiate and receive updates via text — critical for the carrier's non-standard auto demographic where mobile is the primary access channel
- Voice: Amazon Connect contact center routing for phone-initiated claims, with Amazon Lex handling intent recognition and data capture before human handoff
Every channel feeds the same downstream pipeline — a unified claims record in Amazon DynamoDB, with consistent data structure regardless of how the claim was initiated.
Once FNOL is submitted, the platform automatically processes all supporting documents without human intervention — extracting, validating, and enriching the claims record before any adjuster reviews it.
- Driver's license validation: Documents uploaded via the web portal trigger an AWS Lambda function that invokes Amazon Textract to extract and validate identity information, cross-referencing against policy records
- Vehicle damage assessment: Accident images are analyzed by Amazon Bedrock using Amazon Nova Pro, which assesses damage severity, estimates repair and replacement costs, and generates a structured damage report — all in seconds, before the adjuster opens the file
- Document summarization: Amazon Bedrock processes unstructured documents such as police reports and claim forms using NLP, extracting key facts and surfacing discrepancies between submitted documents and policy records
This automation directly addresses the industry benchmark showing AI-powered document processing delivers up to 50% cost reduction on document-driven tasks — applications, submissions, and forms that previously required manual review.
Before any claim reaches an adjuster's queue, it passes through an automated triage and fraud scoring layer that classifies severity, routes by complexity, and flags suspicious patterns.
- Automated triage: Claims are classified by severity, complexity, and priority using generative AI — routing straightforward low-complexity claims toward accelerated processing while escalating high-severity or ambiguous cases to senior adjusters
- Fraud detection: Amazon Bedrock and machine learning algorithms analyze incoming claim patterns against historical data, detecting anomalies including duplicate submissions, inflated damage assessments, and behavioral patterns consistent with organized fraud rings
- Duplicate claim detection: Cross-reference logic identifies multiple submissions for the same incident — a common pattern in the non-standard auto segment — before adjudication resources are invested
Industry data from Deloitte identifies fraud detection as one of the highest-ROI applications of AI in insurance, with AI-powered fraud tools projected to grow from a $4 billion market in 2023 to $32 billion by 2032. P&C carriers deploying real-time fraud scoring at intake — rather than post-adjudication — avoid investing processing effort in claims that should never have entered the adjudication queue.
Adjusters access a purpose-built adjudication interface that surfaces everything they need to make a decision — without switching tools, pulling records manually, or reconstructing the claim narrative from disconnected sources.
- Unified adjuster dashboard: All claim details, document analysis results, damage assessments, fraud scores, and policy information in a single view — eliminating the manual data gathering that consumes up to 40% of adjuster time
- GenAI Knowledge Assistant: Adjusters can query a Bedrock-powered knowledge base using natural language — asking questions like "What is the average collision repair cost for this vehicle class?" or "What does the policy cover for uninsured motorist claims?" — and receive grounded, document-backed answers via RAG (Retrieval-Augmented Generation) with Amazon OpenSearch Serverless as the vector database
- Decision workflow: Approve, reject, or request additional information — all actions captured with full audit trail, timestamps, and adjuster attribution in DynamoDB
AI-assisted adjudication aligns with McKinsey's finding that domain-level AI transformation in claims delivers a 3–5% improvement in claims accuracy — and with industry benchmarks showing AI has compressed standard claims processing cycles by up to 50% when automation handles triage and document preparation.
Once an adjuster decision is recorded, the platform handles all downstream communication and fulfillment automatically.
- Decision notification: AWS Lambda picks adjuster decisions from Amazon SQS and immediately notifies the claimant via Amazon Connect — SMS or voice — with the outcome and next steps
- Payment processing: Stripe integration enables direct settlement payment for approved claims without manual accounts payable intervention
- Policy admin system sync: Lambda functions running in a VPC update downstream systems including the carrier's existing policy administration platform, ensuring claims records stay synchronized without manual re-entry
- Third-party claims system integration: Guidewire and Socotra integration endpoints support carriers who need claims data to flow into enterprise claims management systems for further processing
Every action taken within the platform — from FNOL submission to adjuster decision — is captured in an immutable, queryable audit trail designed to meet the requirements of the NAIC AI Model Bulletin and state-level AI governance frameworks now taking effect across US insurance markets.
- Full decision audit trail: Every AI-assisted decision is logged with inputs, model version, outputs, and adjuster override history in DynamoDB — providing the versioned, explainable record that regulators now require
- AWS WAF + Cognito: All web interfaces protected with WAF rules and authenticated via Cognito, ensuring only authorized users access claims data and all access events are logged
- CloudWatch observability: End-to-end observability across the entire claims pipeline — Lambda execution, Bedrock inference, SQS queue depth, Connect interaction volumes — enabling proactive performance management and compliance reporting
This infrastructure directly addresses the compliance gap identified in the research: most carriers using AI in claims lack versioned model audit logs, explainability documentation per decision, and the structured evidence needed to produce a regulator-ready package during a market conduct examination.
“This engagement required us to balance speed with compliance rigor. We deployed infrastructure-as-code from day one, automated evidence collection across the environment, and delivered a production-ready architecture that passed security review on the first attempt.”
The results speak for themselves.
QyrosCloud transformed a manual, single-channel, inbox-dependent claims process into a production-grade omnichannel platform with generative AI embedded at every stage of the claims lifecycle. The carrier moved from zero automation to a system where the majority of routine claims — FNOL intake, document validation, fraud scoring, damage assessment — are handled without human intervention before the adjuster ever opens the file. The platform doesn't just accelerate existing workflows. It fundamentally changes the economics of claims handling, the quality of the claimant experience, and the carrier's posture against both fraud and regulatory scrutiny.
A non-standard auto insurance carrier serving high-risk and cross-border markets, processing first notice of loss and claims across multiple US states and Mexico.
Related stories.
Ready for results like these?
Let's talk about your AWS environment.
Book a discovery call