From 20 to 2 Hours: AI-Powered Order Processing

Summary
A CNC contract manufacturer in Central Switzerland partnered with iosys to replace manual order entry with an AI-powered automation solution. The system automatically extracts order data from PDF emails and transfers it to the ERP. The result: a 90% automation rate, reducing weekly processing time from 20 to 2 hours – and a CEO who can focus on the core business again.
Starting Point
The company is a typical Swiss manufacturing SME: precision CNC machining for industrial customers, over 50 customer orders per week, and a historically grown landscape of ERP software and Excel spreadsheets for job tickets, material tracking, and order planning.
The problem: Every order arrived via email as a PDF – and had to be manually entered into the ERP system.
"Can't we automate this in a simple way?"
The Challenge
The manual process tied up valuable resources:
- 20 hours per week for data entry – by the CEO himself
- Error-prone repetitive work
- Delays during order peaks
- Knowledge in heads instead of systems
The typical workflow: Open email, read PDF, switch to ERP, enter article number, quantity, and delivery date, check inventory, send confirmation. With 50+ orders per week, this added up quickly.
The Solution
iosys developed a tailored automation solution together with the manufacturer:
Email inbox (PDF)fetchAI extraction (API)extractERP integrationtransferPlanningschedule
Technical components:
- n8n Workflow Engine (Open Source) – self-hosted, no vendor lock-in
- Outlook Inbox Monitoring – automatic retrieval of new orders
- Large Language Model (LLM) – for precise PDF extraction
- Few-Shot Learning – system learns customer-specific formats
- ERP Cloud API – direct integration with the existing ERP system
What employees still do:
- Review exceptions and edge cases
- Create new customers or articles
- Spot checks
"Automation also means rethinking processes – not just digitizing 1:1."
The Approach
How iosys worked together with the customer:
- Problem analysis: Where does most time go?
- Feasibility check: Is AI extraction from PDFs realistic?
- Pilot phase: Small tests with real orders
- Iterative development: Step-by-step improvement
The decision criteria were clear: No big bang, but gradual progress. Involve employees early. Quality over speed.
The hurdles:
- ERP API integration was more complex than expected
- Edge cases with special formats required rework
- Cloud migration of the ERP was a separate sub-project
The Results
| Metric | Before | After |
|---|---|---|
| Time per week | 20 hours | 2 hours |
| Automation rate | 0% | 90% |
| Manual steps | 6 per order | Only for exceptions |
| Project Duration | Investment |
|---|---|
| 3 months to the result, + 2 months ERP migration | approx. 20k CHF* |
*including ERP cloud migration
ROI: Payback within months through saved labor time.
What Worked
- Iterative approach: Building trust through small wins
- Open collaboration: Regular exchange, short feedback loops
- AI accuracy: The extraction works reliably
- Quick acceptance: Even the owner was convinced quickly
"Das isch ja genial!" (That's just brilliant!)
Outlook
The project continues. Planned extensions:
- Automatic supplier confirmations preparation
- Business rules validation – validate business logic (with Human in the Loop)
- On-premises language model – for maximum data security
Facing Similar Challenges?
Whether it's order processing, production data, or ERP integration – iosys helps manufacturing SMEs unlock operational improvements without enterprise budgets.
Learn more about our AI-powered order processing service.