Extracting Invoices From WhatsApp Photos With AI Vision (Apps Script + Google Sheets)

Short summary

A practical pipeline for extracting invoice data from WhatsApp photos using AI vision models (Gemini, Claude, GPT-4o) routed through Apps Script into Google Sheets. A confidence_score threshold auto-appends high-confidence extractions, flags borderline cases for human review, and asks drivers to re-photo low-confidence ones. At ~500 receipts/week the setup costs $40–100/month versus $2,000–4,000/month in manual labor, with 92–97% accuracy on legible receipts.

• •Pipeline: WhatsApp image → Apps Script → vision model → JSON with confidence_score → Google Sheet
• •Cost: ~$0.005–0.02 per receipt; 92–97% accuracy on legible receipts, 75–85% on damaged
• •Confidence routing: >90 auto-append, 70–90 human review, <70 re-photo request

Generated with AI, which can make mistakes.