GenAI-Powered Invoice Processing Solution

FinanzChef24

Success Story

About FinanzChef24

FinanzChef24 is a German digital insurance broker for self-employed professionals and small businesses. The company helps customers compare business insurance offers across many providers, understand policy details and complete coverage entirely online. With a strong focus on clarity, speed and reliable support, FinanzChef24 reduces the complexity typically associated with commercial insurance and enables customers to make well-informed decisions.

To keep internal financial operations efficient as volumes increased, FinanzChef24 looked to modernize invoice handling across multiple insurers and document formats.

“Auvaria helped us turn a slow, manual invoice process into a structured, automated pipeline. We can already extract commission-relevant data consistently across different providers’ formats and adding a new insurer looks a lot simpler than before. It’s a promising step forward for our finance team.”

Vlad-Gheorghe Barboni
Co-CEO / CTO

Project Overview

Branch

Financial Services / Insurance Brokerage

Topics

GenAI document understanding + Invoice data extraction + Structured JSON output + Serverless processing

Project Duration

4 Months

Customer Inquiry

How can we reliably extract all required financial fields from invoices across multiple insurance providers and formats, while scaling processing volume and keeping results consistent for downstream commission calculations?

Project at a Glance

1

Challenge

Invoices arrived in many provider-specific layouts, making manual extraction slow, error-prone and difficult to scale.

2

Solution

A serverless pipeline on AWS that uses Amazon Bedrock with Claude 3.7 Sonnet and provider-specific prompts to produce consistent, structured JSON.

3

Outcome

Faster invoice handling with improved consistency of extracted fields and easier onboarding of additional providers.

Objectives

  • Automate Structured Extraction
    Replace manual copy-paste work by extracting policy and commission-relevant fields into a standard JSON schema.
  • Handle Provider Variability
    Support different invoice layouts and terminology across insurers without a fragile one-size-fits-all approach.
  • Improve Downstream Reliability
    Deliver consistent, validated outputs suitable for commission calculation and internal financial processing.
  • Scale Cost-Effectively
    Use a serverless architecture that can handle growing invoice volume without heavy operational overhead.

Challenges

  • Highly Variable Document Layouts
    Each insurance provider used different structures, labels, and field placements, requiring robust, context-aware extraction.
  • Context-Heavy Terminology
    Insurance documents contain domain-specific wording where meaning depends on surrounding context, not just keywords.
  • Quality and Consistency Requirements
    Small extraction errors can propagate into scoring and commission calculation, increasing rework and audit effort.
  • Operational Bottlenecks
    Manual processing created throughput constraints and inconsistent results across teams and time periods.

How Our Method Drives Results

Auvaria Approach

Serverless Orchestration with AWS Lambda

Implemented event-driven processing that reacts to new invoice uploads and routes documents through extraction steps.

Secure Document Storage on Amazon S3

Stored invoices in Amazon S3, organized in provider-specific paths to support separation of concerns and onboarding.

LLM-Based Extraction with Amazon Bedrock + Claude 3.7 Sonnet

Used Amazon Bedrock with Claude 3.7 Sonnet to interpret PDF content and extract complex, context-dependent financial fields.

Provider-Specific Prompt Modules

Developed modular prompts per insurer to align extraction with provider vocabulary and layout, improving accuracy versus generic prompting.

Evaluation-Driven Model Selection

Compared multiple models using a model-as-a-judge setup and selected Claude 3.7 Sonnet for strong faithfulness and structured output performance.

Success Story

The

Results

  • Faster Invoice Processing
    Reduced manual effort by automating extraction of the core fields needed for financial workflows.
  • More Consistent Outputs
    Standardized JSON results improved reliability for downstream commission calculation and reduced formatting-related rework.
  • Easier Provider Onboarding
    Modular provider-specific components simplified adding new insurers without redesigning the entire pipeline.
  • Scalable Operations
    Serverless execution supported variable invoice volumes while keeping operations lightweight.
  • Clear Path for Enhancements
    Established a foundation for confidence scoring, human-in-the-loop validation and continuous prompt/model iteration.

Sascha Alker

Senior Account Executive

Ready to transform your business?

Let's turn your vision into reality