Overview
Sirma has extensive expertise in implementing multi-technology AI solutions to automate complex insurance processes, utilizing OCR, machine learning, and workflow integration. Our team delivered a scalable, efficient, and impactful digital solution for PetsBest, a leading provider of veterinary insurance. We assisted in automating their claims processing workflow, transforming what was once a manual and labour-intensive task into a streamlined, AI-powered system. The solution employs optical character recognition (OCR) and machine learning to automatically classify claims from veterinary invoice documents, enhancing the entire back-office process and improving operational efficiency.
The Challenge
Pets Best offers pet insurance and wellness plans in the USA. The company aims to provide affordable, comprehensive animal healthcare with flexible coverage and an easy claims process. The clinic faced significant challenges in managing the high volume of insurance claims submitted through paper or image-based veterinary invoices, which were frequently faxed in from clinics. The manual review process was not only time-consuming but also prone to errors, limiting the company’s ability to efficiently scale its claim processing. The objective was to automate the extraction, classification, and initial decision-making for claims while accurately handling various unstructured invoice formats and complex veterinary procedure codes.
The Project Scope
Sirma was engaged to develop and deploy an end-to-end automated claims classification system that would:
- Utilize OCR technology to digitize veterinary invoices sent via fax or image upload;
- Parse invoices down to individual line items representing specific veterinary procedures;
- Classify each line item by procedure type and insurance coverage category;
- Make an initial automated decision to approve claim payment or route it for human review;
- Seamlessly integrate with PetsBest’s existing claims management system;
- Support continuous learning to improve classification accuracy over time.
The Solution
Sirma created an ML-fed pipeline with advanced optical character recognition (OCR) to extract text and numeric data from various invoice formats. The system processes scanned invoices into detailed line items and uses supervised classification models to categorize and assess coverage for each item. A decision engine evaluates claims based on business rules and model outputs, deciding whether to approve them automatically or escalate them for human review.
The integration of human review with automated decision logic allows the system to route claims or bids to human reviewers when uncertainties or low-confidence classifications are detected. This streamlined process ensures that flagged cases are efficiently evaluated within the existing workflow. The integration leverages:
- Rule-based triggers combined with machine learning prediction confidence to distinguish clearly classified items from uncertain ones;
- Automated prioritization and queuing mechanisms that feed flagged items into the human review pipeline promptly;
- Feedback loops where human review outcomes are used to retrain and fine-tune ML models, improving future classification accuracy and reducing manual escalation over time;
- User interfaces and workflow tools that allow human experts to view automated annotations, override decisions if needed, and submit final decisions back into the system.
Results
- The automated system achieved a 300% increase in claim processing productivity, reducing manual workload drastically;
- Claims that do not require human review are processed faster, accelerating reimbursement cycles and improving customer satisfaction;
- The solution was acquired and is now owned by Synchrony, where it continues to support image-based invoices, proving its extensibility and lasting business value;
- Improved classification accuracy and decision consistency reduced errors and potential fraud exposure;
- PetsBest realized operational cost savings and improved scalability for growing claim volumes.
Technologies
- Advanced Optical Character Recognition (OCR) engines tailored to handle veterinary invoice formats;
- Custom machine learning classifiers for line-item classification based on veterinary procedure and coverage codes;
- Supervised learning models with continuous training on claim outcome feedback;
- Automated decision-making logic embedded in the claims processing workflow;
- Integration components to embed AI classification seamlessly into PetsBest’s claims management systems;
- Support for multi-channel input, including fax, email, and image uploads.
Sirma’s Partnership with the client
In this project*, our team created a hybrid AI-human decision-making ecosystem that maximises automation efficiency while preserving expert oversight and control for complex or ambiguous cases. It enhances productivity without sacrificing accuracy or risk management, typical of intelligent enterprise AI transformations led by Sirma.
*The solution, now, is owned by Synchrony.
