Invoice Extraction: Quality & Cost
Compare four methods on 100 synthetic invoices using recorded quality, latency, API costs, and explicit review-cost assumptions.
Explore the workflow
Compare four tested invoice-extraction implementations on 100 synthetic invoices. Explore saved evaluation scores, recorded API costs and mean latency, run live extraction, and model review costs with explicit assumptions. Historical results include a flagged Textract address-parser limitation; live outputs are unscored.
- Regex + per-vendor templates
- AWS Textract AnalyzeExpense
- Claude Sonnet vision extraction
- Hybrid Textract + Claude reconciliation
- Live per-document extraction with all four methods