Manual data entry → 90% faster invoice processing

The problem

A mid-market finance team was rekeying invoices from PDFs into NetSuite — eight FTEs, daily, with a 3% error rate.

The solution

An extraction pipeline using GPT-4 Vision + structured outputs, validated against a deterministic rules engine, with a human-in-the-loop UI for low-confidence rows.

Approach

We replaced manual data entry with a model + rules hybrid: the LLM handles messy formats, and a deterministic validator enforces invariants the business already trusts.

Detail

Low-confidence rows route to a tight human-in-the-loop UI instead of being silently wrong. Every decision is logged, so audit and finance both stay happy.

Sample evaluation harness

// evals/extraction.eval.ts
import { runEval } from "./harness";
import { invoiceCases } from "./cases";

await runEval({
  name: "invoice-extraction",
  cases: invoiceCases,
  judge: "strict-json-schema",
  passThreshold: 0.95,
});