Automating Document Processing to Strengthen Service Quality

Challenge

Solidaris CSD processes a large volume of legal documents submitted by beneficiaries, which serve to justify the in-home care services delivered by its teams. While this information is essential, encoding it manually into internal systems represents a significant use of staff time. To help teams focus on higher-value activities and maintain quality of care, Solidaris CSD launched a project to automate the encoding process. The aim is to accelerate processing, improve data reliability, and allow staff to dedicate more time to beneficiary support and service coordination.

Nature of collaboration

Sagacify collaborated with Solidaris CSD on a proof-of-concept to design and deliver an AI solution to automate the processing of service records from care teams. Solidaris defined the classification structure, provided representative documents, and clarified the data requirements for integration with their internal system. Using this input, Sagacify annotated the documents to create training data, then led the end-to-end implementation—designing the document pipeline using Skwiz, training models for classification and extraction, and aligning the outputs for integration with IRIS, Solidaris' internal document management system.

Solution

The system was built using Skwiz, Sagacify’s AI platform for document automation. It processes service records submitted by care teams—often as scanned PDFs or photographs—through a structured pipeline:

Document classification: A supervised model categorizes each document by its type and associated service, based on Solidaris’ internal classification system.
Information extraction: Key fields such as the beneficiary’s name, NISS number, address, and case reference are identified and extracted.
Data integration: The extracted data is prepared for upload into IRIS, Solidaris’ internal document management system.

The models were trained on actual documents and refined using annotated examples provided by Solidaris. Their performance is measured using standard accuracy metrics that reflect both precision and completeness across all document categories.

‍