Phase 1: AI model development. Loncom built a registry-based AI architecture with separate, dedicated invoice reading models for each energy supplier. Rather than applying a single generic model across all invoice types, each supplier model is configured with the specific validation and extraction logic needed to handle that supplier’s invoice format accurately. Models were built and refined for SSE (including consolidated invoice formats), E.ON, EDF, Pozitive Energy (including single and consolidated invoices), Crown Gas and Power (including electricity and gas variants), and Corona Energy (electricity and gas). Each model was trained to extract: supplier name, MPAN, invoice number, invoice date, start and end dates, consumption, VAT, CCL, standing charges, NET values and credit note handling. An OCR layer was implemented for scanned or image-based invoices. Duplicate detection and flagging was built to prevent the same invoice from being processed twice.

Phase 2: Platform build and deployment. A full web platform was built around the AI engine: a secure frontend with login, bulk invoice upload with PDF preview, a processed invoice review interface and export functionality. The platform was deployed on AWS with Git Actions CI/CD for automated deployments. A background worker was implemented to handle invoice processing asynchronously, preventing timeouts on large batches. Supplier name alignment was built to match extracted supplier names to the client’s existing property management system (Qube). MySQL database infrastructure was configured and later migrated to MySQL 8.0 for performance and supportability. Enhanced logging was implemented to assist with debugging AI model failures. The platform was deployed to Signature Energy’s own domain and separate user logins were created for each team member.

Phase 3: Retainer and continuous AI refinement. Invoice reading is an ongoing accuracy problem: supplier invoice formats change, edge cases emerge and new invoice types (consolidated invoices, credit notes, multi-meter invoices) require new logic. Loncom’s retainer covers continuous AI model refinement, bug fixes across all supplier models, platform improvements, and the development of additional capabilities including a CRM layer and further AI reading improvements. The platform remains in active development with 42 features currently in progress or under review.