Automate document processing with Snowflake Document AI and DocProc
How to turn incoming invoices, receipts and delivery notes into structured rows in your Snowflake account — without ever moving the documents out of your tenant.
Why Snowflake-native
Most document automation tools ship your PDFs to a third-party SaaS. With Snowflake Document AI plus DocProc, the model runs inside your Snowflake account: the source files stay in your stage, extractions land in your tables, and your existing RBAC and audit trail apply. There is no extra data residency surface to negotiate with security.
Architecture in one paragraph
DocProc picks up incoming documents from an email mailbox or a watched stage, normalises them, and calls a Snowflake Document AI model you have trained on your own invoice and receipt samples. The structured output is written back to a Snowflake table, optionally routed to a Human in the Loop queue for low-confidence fields, and then handed to your ERP — Abacus, SAP, Microsoft Dynamics — over its standard API.
Step 1 — Train a Document AI model on your own samples
In Snowsight, create a Document AI build, upload 20–50 sample documents per type (Eingangsrechnung, Quittung, Lieferschein), and label the fields you care about: supplier, invoice number, date, VAT, IBAN, line items. Snowflake produces a model you can call from SQL.
Step 2 — Wire DocProc to the model
Point DocProc at the Snowflake stage that holds the incoming PDFs and at the PREDICT function of your Document AI build. DocProc handles classification (which model to call per document type), retries, and the writeback into your tables.
Step 3 — Route low-confidence fields to a human
Every extracted field carries a confidence score. DocProc's Human in the Loop UI surfaces only the fields below your threshold; the rest flow straight through to the ERP. Reviewer corrections feed back into the next training round, so the model improves on your data.
Step 4 — Hand off to the ERP
Once a document is approved, DocProc posts a creditor draft to your ERP and files the original under a deterministic name (YYYYMMDD_SUPPLIER_INVOICE.pdf) so finance can always trace a booking back to the source.
What you get
- Documents stay inside your Snowflake account — no external SaaS to audit.
- Field-level confidence and a Human in the Loop for the long tail.
- ERP writeback for Abacus, SAP, Dynamics and friends.
- Swiss-made, GDPR/FADP-aligned, run from Zurich.
Want to see this on your own invoices? Get in touch.