Invoice data extraction that reconciles itself

What it does

• Reads PDFs, scans, photos, Word, Excel and email attachments
• Extracts header fields and every line item
• Checks sum of lines = stated total, tax rates and dates, in code
• Matches invoices to purchase orders and delivery dockets (two- and three-way matching)
• Reconciles supplier statements against your ledger: missing invoices, keying errors, duplicates
• Marks each invoice pay, hold or escalate, with the reason written down

Measured results

Tests on generated documents with exact known answers. The statement reconciliation needed a 64k-token answer from Opus; Yellowjacket's workers read one statement at a time and code does the matching.

Why accounts payable teams use it

An AI that reads invoices is only useful if you can trust its numbers. Yellowjacket never asks a model to add up: models read, code calculates, and a reading that doesn't reconcile is read again by a stronger model. Every invoice has a record of which model read it and which checks it passed.

How it works

1. A top model plans the job once. Claude Opus looks at a few samples and writes the instructions, the checks and the arithmetic. You pay for that once per kind of job, and the plan is kept for next time.
2. Cheap models do the reading. A short tryout picks the cheapest open models that pass the job's checks; they read every document. Sums, dates and comparisons are done in code, not by a model.
3. Every answer is checked. Totals have to reconcile and two different models have to agree. When they don't, a stronger model reads it again; only real disagreements go back to the top model.
4. You get a receipt. Before the job you see what Claude, ChatGPT and Gemini would each charge alone; after it, what each document cost and which checks it passed.

Related