5 weeks to move from manual underwriting intake to AI-driven submission automation on AWS

Insurance underwriting teams struggle with unstructured submissions, manual triage, and slow turnaround times. Traditional intake workflows rely heavily on human effort, leading to delays, errors, and poor visibility.

By implementing a generative AI-powered, AWS-native automation system, Blue Komodo transformed submission intake into a fast, structured, and decision-ready pipeline.

Customer overview

Blue Komodo is a technology consulting firm focused on applying AI to solve operational bottlenecks in document-heavy industries like insurance and finance. They specialize in building practical AI systems that automate repetitive workflows and improve decision-making. For this engagement, they partnered with Applify to validate how generative AI could streamline insurance submission intake and underwriting workflows.

“Applify helped us move beyond experimentation to a real, production-ready Gen AI solution. We saw a 70 to 80% reduction in manual effort while improving consistency and turnaround time.”

Shane Murphy, Founder, Blue Komodo

Challenges | Manual intake is slowing down underwriting teams

Insurance submissions arrive in inconsistent formats across emails, PDFs, Word files, and spreadsheets. Without structured intake systems, underwriters are forced to manually interpret and route each submission.

This creates delays, increases error rates, and limits the ability to scale operations efficiently.

Manual processing overload

Underwriters spent significant time reading emails and extracting data from unstructured documents across new business, renewals, endorsements, and broker of record transfers.

Inconsistent data formats

Submissions varied across brokers and carriers, making standardization difficult without human intervention.

High error rates and delays

Manual data entry led to inaccuracies and slower turnaround times, impacting SLAs and broker experience.

Lack of operational visibility

No clear insight into submission volumes, processing times, exception rates, or audit trails required for regulatory compliance. 

Solution | Generative AI-powered submission intake and underwriting automation

We built an end-to-end generative AI-powered workflow on AWS that automates submission ingestion, data extraction, classification, and routing. The system transforms unstructured inputs into structured, validated data that is ready for underwriting decisions within seconds.

Generative AI-powered infrastructure on AWS

The solution uses a serverless, secure-by-design architecture powered by Amazon Bedrock to process submissions at scale while maintaining accuracy and flexibility.

Intelligent document processing

Automatically reads and understands emails and attachments across formats using Amazon Bedrock and Anthropic's Claude foundation model. Domain-specific, few-shot prompting guides the model to accurately interpret varying submission structures, removing the need for underwriters to manually process documents.

Automated data extraction

Extracts key fields like insured details, coverage types, policy limits, effective dates, loss history, and broker information using AWS Textract and Amazon Bedrock, reducing manual entry and minimizing errors.

Intent classification

Identifies whether a submission is new business, renewal, endorsement, or broker of record transfer using Amazon Bedrock, enabling faster and more consistent routing decisions, even in edge cases where submission intent is ambiguous.

Carrier-specific validation

Applies configurable business rules per carrier and product line to validate data completeness, flag inconsistencies, and surface potential quality issues before routing.

Automated broker correspondence

Generates contextual follow-up emails to brokers requesting missing information or clarification on flagged exceptions, reducing manual follow-up effort and accelerating resolution.

Intelligent submission routing

Assigns validated submissions to the right underwriters based on product expertise, geographic territory, workload balancing, and submission complexity, ensuring efficient allocation without manual intervention.

Event-driven ingestion

Captures and processes incoming submissions in real time using AWS Lambda and event-based triggers, eliminating delays caused by manual intake.

Serverless workflow orchestration

Coordinates the entire intake lifecycle using AWS Step Functions, ensuring scalable and reliable execution without infrastructure management.

Exception handling workflows

Routes low-confidence or incomplete submissions for human review using AWS Step Functions and Lambda, ensuring accuracy without slowing overall throughput.

Real-time monitoring

Tracks submission volumes, processing times, and exception rates through AWS-native monitoring tools. Complete audit trails provide full operational visibility and support regulatory compliance requirements.

Success metrics

With generative AI at the core, submission intake shifted from manual processing to intelligent automation, improving speed, accuracy, and scalability across the entire underwriting pipeline.

• 80% automation of standard submissions: Most submissions were processed end-to-end without human intervention.
• 78% reduction in ingestion time: Processing time reduced from hours to seconds for structured output generation.
• 30% productivity improvement: Underwriters spent more time on risk assessment instead of manual data entry.
• Production-ready in 5 weeks: From POC to a deployable system within a short sprint cycle.
• Scalable to high-volume operations: Architecture designed to handle up to 100,000 submissions per month with strong cost efficiency.