Automate Invoice Reconciliation with a Custom AI System

Most mid-sized accounting firms rely on the AP automation features within QuickBooks Online or dedicated tools like Bill.com. These tools work well for standardized invoices from large vendors but falter with complexity. QuickBooks' OCR is template-based; it fails when a small subcontractor sends a multi-page PDF with a slightly different layout than last month, forcing a junior accountant to revert to manual data entry.

Bill.com offers more advanced workflows, but its matching logic is rigid. Consider a firm with a construction client. The client’s purchase order says “100 ft. copper pipe” but the vendor invoice lists “Copper Piping - 100 feet.” A standard system flags this as a mismatch, requiring manual override. When 20% of invoices have minor variations like this, the 'automation' creates more review work than it saves.

Here is a common scenario. An accountant is reconciling invoices for a property management client. They receive a 15-page invoice from a general contractor with line items for materials, subcontractor labor, and equipment rentals, all billed to different properties. The accountant must manually split the single invoice into dozens of journal entries, allocating costs to the correct property and expense account. This process takes 45 minutes for one complex invoice and is highly prone to error.

The structural problem is that off-the-shelf tools are built for the most common denominator. Their data models are fixed, and their AI is an add-on feature, not the core architecture. You cannot teach them your client's specific chart of accounts or a vendor's unique invoicing quirks. Your team is forced to work around the software's limitations, creating a permanent ceiling on efficiency.

The first step is a discovery audit. Syntora reviews a sample of 50-100 of your firm’s most challenging vendor invoices across several clients. We identify the different formats, the required data fields (GL codes, client matter IDs), and the specific matching rules. This audit produces a clear scope document outlining the data extraction and validation logic required for a custom system.

The technical approach uses a modern AI stack built for flexibility. A Python service uses the Claude API for intelligent data extraction. Unlike rigid OCR, this can interpret context from non-standard layouts, handwritten notes, and multi-page documents. The extracted data is structured into JSON and passed to a FastAPI application that applies your firm's unique matching and validation rules against data from your accounting system. All processed data and an audit trail are stored in a Supabase PostgreSQL database.

The delivered system is a simple, secure web portal where your team can upload invoices in bulk. Within seconds, the system displays the extracted data, its match confidence against POs or ledger entries, and any flagged exceptions. Approved invoices are pushed directly into your accounting software via its API. You receive the full source code, a maintenance runbook, and a system that runs in your own secure cloud environment.