Stop Stockouts: Custom AI for Urgent Care Inventory

Most clinics start with a general inventory app like Sortly. This requires a nurse to manually scan a barcode every time a box of gloves or a vial is used. During a patient surge, scans are inevitably missed. The system shows 5 boxes in stock, but the shelf is empty, leading to a critical supply stockout during a procedure.

A practice management system's built-in module seems like an improvement, but it has different flaws. It might track items tied to a billable CPT code but completely misses non-billable supplies like gauze or alcohol swabs. Its reordering logic is rigid, often just a simple rule like 'reorder when stock hits 10 units'. This ignores seasonal demand, like the winter spike in flu tests, causing over-stocking in slow months and under-stocking in busy ones.

These off-the-shelf tools fail because they are fundamentally disconnected from real-time clinical activity. They depend on flawed manual entry or simplistic billing triggers. Neither can accurately model the complex, fast-paced consumption patterns of an urgent care environment, making them unreliable for preventing costly stockouts.

Syntora would approach inventory optimization by first conducting a discovery phase to audit your existing data sources and operational workflows. The technical approach would involve connecting directly to your Electronic Health Record (EHR) via its API to extract historical consumption data. We have experience integrating diverse data sources and mapping complex datasets, similar to our work with financial documents, to build the foundational data required for accurate inventory management. The goal here is to establish a reliable, historical dataset of exactly what was consumed and when.

Following data preparation, Syntora would develop a custom consumption forecasting model using Python. This model would be designed to analyze your clinic's specific time-series data, identifying unique seasonal patterns for your inventory items and predicting demand up to 30 days into the future, potentially utilizing libraries like Prophet. The forecast calculations would be automated, for instance, running nightly as an AWS Lambda function, triggered by a CloudWatch event.

A lightweight FastAPI service would be developed to manage the core inventory logic. This service would compare the daily 30-day forecast against current inventory levels, which would be stored in a Supabase database. When predicted consumption is projected to breach a pre-defined safety stock level (typically calculated considering supplier lead time and a configurable buffer), the service would generate a draft purchase order. Pydantic would be incorporated to ensure all generated PO data adheres to a correct and verifiable structure before it is presented for approval.

The delivered system would be deployed on platforms like Vercel and AWS. Operating costs for such custom solutions are typically optimized to be low, often remaining under $100 per month for clinics with substantial patient volumes, depending on the specific scale of data processing and API usage. We would implement structured logging with structlog and configure custom alerts (e.g., via Slack) for critical operational events, such as failed API calls to suppliers or unexpected model deviations. Our engagement includes providing the full source code and documentation, along with knowledge transfer to empower your team with the understanding and tools to maintain and evolve the solution.