Finance ops · 8 months live
Invoice processing — email inbox to ERP, zero manual entry
The problem
Manually creating client records and invoices for every order — pure double data entry, hours each day, 5% error rate from copy-paste mistakes.
First attempt
First attempt used a rules-based extraction engine — broke immediately on non-standard invoice layouts. Claude API handled layout variation reliably from day one.
Workflow diagram · grounded in source
1
Email inbox
Trigger
2
N8n
Orchestration
3
Claude API
AI extraction
4
Postgres
Deduplication
5
ERP API
Output
Reported outcome
200+ invoices/day processed unattended.
Replaced a €900/mo contractor. Setup took 3 days. Running 8 months without a single manual intervention. Error rate dropped from 5% to near zero.
Reported metrics
Time saved4 hrs/day
Volume200 invoices/week
Running sinceAug 2024
Reported stack
Email inboxn8nClaude APIPostgresERP API
Frequently asked questions
What did this team achieve with this AI workflow?
200+ invoices/day processed unattended.
What tools did this team use?
Email inbox, n8n, Claude API, Postgres, ERP API.
What results were reported?
Time saved: 4 hrs/day; Volume: 200 invoices/week; Running since: Aug 2024 (source-reported, not independently verified).
What failed first in this deployment?
First attempt used a rules-based extraction engine — broke immediately on non-standard invoice layouts.
How is this finance ops AI workflow structured?
Email inbox → N8n → Claude API → Postgres → ERP API.