# Order Processing with AI Agents: The Pattern Behind 96% Automation

> Author: Chris Jon Graf (AI Strategist & CEO)
> Updated: 2026-10-05
> URL: https://ai-outsourcing.ch/insights/order-processing-with-ai-agents-the-pattern-behind-96-automation

## Summary

Order processing with AI agents works when you choose the highest-volume, clearly structured process, set a confidence threshold for human-in-the-loop, and secure data quality. ACO, Primavera, and SCHUNCK reach 50–96 percent automation. The pattern transfers to SMEs—without an in-house AI team, in weeks rather than months.

## The pattern: Start with the biggest pile, not the hardest case

Three European companies have taken order or claims processing to 50–96 percent automation—not with exotic technology, but with a repeatable pattern. You don't start with the hardest single case; you start with the highest-volume, clearly structured process. AI agents extract from emails, PDFs, and Excel lists; RPA handles structured bulk data. Humans stay in the loop when confidence is low. This approach requires leadership commitment—see [how mid-market companies make AI a management priority](https://www.ki-podcast.ch/ki-im-mittelstand-management-thema-nicht-it-projekt).

### ACO: From manual data entry to project work

**>50%** — Automation rate in ACO's order processing

ACO, a building materials manufacturer with 5,400 employees and €1.2 billion in revenue, processes tens of thousands of orders per year. With the Workist AI agent, per-document processing time fell from 3–5 minutes to a few seconds; quotes are generated in an average of 4.5 hours instead of days. More than 50 percent of order processing runs automatically. The decisive factor is the confidence threshold: ACO itself determines from which confidence value an order goes directly to SAP without human review.

> 90% of the work is your own organization, above all data quality.
>
> — Rouven Gehrmann, ACO

The 90 employees in customer sales service were not cut; they moved to higher-value tasks: project tracking, CRM maintenance, and appointment scheduling for field sales. Nobody wants to just type data anymore. Even for hundreds of pages of specifications, a dedicated assistant supports the team.

### Primavera: 55 percent fully automatic in natural cosmetics distribution

**55%** — Fully automatic orders at Primavera

Primavera achieves over 80 percent time savings in order entry; 55 percent of all orders are processed fully automatically and without errors. The path: email inbox, Workist AI agent, then Sage-100 ERP. In case of deviations, the agent asks a human for help—human-in-the-loop as a learning mechanism. User acceptance is high because the agent learns quickly. The goal is the stretch from email to logistics without sales or IT having to intervene.

### SCHUNCK: 96 percent dark processing as a growth lever

**96%** — Dark processing for transport registrations at SCHUNCK

SCHUNCK GROUP, an industrial insurer in the Ecclesia Group, processed 160,000 emails in 18 months using AI-supported extraction. 96 percent of the 40,000 transport registrations run dark; 93 percent of claims notifications are created fully automatically. The trick: don't automate complex single-case underwriting, but the transactional high-volume layer. AI taps unstructured email input, RPA robots harmonize Excel lists. Unit costs fall as volume rises.

> Start with the biggest pile of transactions.
>
> — SCHUNCK GROUP

## Five transferable principles for SMEs

- Choose the highest-volume, clearly structured process—not the most complex one.
- Introduce human-in-the-loop with a company-defined confidence threshold.
- Put data quality before technology—it determines the automation rate.
- Combine AI for unstructured input with RPA for structured bulk data.
- Move employees into higher-value roles instead of cutting jobs.
- Clarify ERP integration from the start—SAP, Sage, or similar.

The strategic framing matters: AI in the mid-market is a leadership topic, not an IT project.

## Your first step

You don't need your own AI team. Specialized providers deliver pretrained agents; entry takes weeks, not months. The ideal first process is an email-based order or purchase order intake that does not arrive via EDI. In Switzerland, also check data sovereignty and revDSG-compliant processing.

How to avoid the three biggest strategic mistakes in global competition is explained by the [Swiss AI Podcast](https://www.ki-podcast.ch/ki-standort-schweiz-kmu-strategie-und-globales-rennen).

> **Concrete starting point**
>
> Choose a process with a substantial transaction volume, a clear data source such as email, and an existing ERP interface. Everything else is a matter for discussion—and exactly where we support you.

## FAQ

### How high can automation go in order processing?

The case studies range from 50 to 96 percent. SCHUNCK achieves 96 percent dark processing for transport registrations. The exact rate depends on data quality, confidence thresholds, and process structure.

### Which process is the best starting point for an SME?

Email-based order or purchase order intake that does not arrive via EDI. It has high volume and clear structure, and can be automated in weeks without an in-house AI team.

### Do we need our own AI team?

No. Specialized providers deliver pretrained agents and handle integration. The organizational groundwork—data quality, process selection, confidence thresholds—remains with you.

### What does human-in-the-loop mean in practice?

The AI agent only processes orders whose confidence score exceeds your defined threshold. When unsure, it asks a human. This keeps control with the company while the agent learns continuously.

### How quickly can we implement this?

With focused scope and an existing ERP interface, a pilot is feasible within weeks, not months.

### What role does RPA play alongside AI?

AI handles unstructured input like emails and PDFs. RPA handles structured bulk data like Excel lists. The combination reduces unit costs as volume grows.

## Sources

- [ACO nutzt die KI-Agenten von Workist im Angebots- und Bestellwesen](https://www.workist.com/success-stories/aco)
- [Primavera & Workist – Über 80% Zeitersparnis in der Auftragserfassung](https://www.workist.com/success-stories/primavera)
- [96 Prozent dunkel: Wie SCHUNCK die Grenze der Automatisierung clever verschiebt](https://www.versicherungstech-magazin.de/96-prozent-dunkel-wie-schunck-die-grenze-der-automatisierung-clever-verschiebt/)
- [Mister Spex automatisiert Service Calls mit KI-Agenten](https://www.cognigy.com/de/case-study/mister-spex)