Unnamed Healthcare SaaS Company

A mid-market healthcare SaaS company managing insurance appeals and grievances workflows faced compounding operational failures rooted in manual case creation. In property and casualty contexts, appeals and grievances carry strict regulatory timelines and documentation requirements — delays translate directly into compliance exposure and degraded policyholder experience. Operators were manually reviewing, categorizing, and keying data from unstructured claim documents, creating bottlenecks that scaled poorly with volume. The consequences were measurable: high operational costs, elevated error rates, risk of critical data loss during handoffs, and an inability to process claims at the speed the business required.

Harbinger's InsurTech team developed an AI and OCR-based automation platform targeting the appeals and grievances case creation pipeline. The core of the solution combined pre-trained NLP models — used for document classification and entity extraction — with custom-trained models built around the client's specific case taxonomy and document formats. OCR handled ingestion of unstructured claim documents, converting them into machine-readable input for the NLP layer. The pipeline automatically categorized incoming documents, extracted relevant insurance data fields, and generated structured case records, allowing operators to review and finalize cases rather than build them from scratch. Integration with the client's existing case management workflows ensured operators experienced a streamlined handoff rather than a parallel system.

The automated pipeline delivered 90% accuracy on case type classification — the highest-stakes determination in the appeals and grievances workflow — and 80% overall accuracy across all insurance data fields extracted and generated. These benchmarks represent a meaningful reduction in the manual correction burden that previously consumed operator time. Qualitatively, the client reported:

• Significant time savings per case for frontline operators
• Reduced risk of data loss during document-to-case handoffs
• Lower error rates compared to the prior manual process
• Improved ability to scale claims volume without proportional headcount increases

• Define accuracy benchmarks for each data category — not just overall — before deployment; case type accuracy and field-level accuracy can diverge significantly and carry different operational consequences.
• Combining pre-trained NLP models with custom-trained models allows teams to leverage general language understanding while still capturing domain-specific terminology and document structures common in insurance claims.
• OCR quality is a ceiling on downstream NLP performance; invest in document pre-processing before assuming model accuracy is the limiting factor.
• Automating case creation rather than case review shifts operator effort to higher-value exception handling, which improves both throughput and job quality.
• In regulated healthcare insurance environments, automation must account for audit trails and data lineage from document ingestion through case generation.