Live Shopfloor Data from a Legacy ERP

Summary

DUAP AG, a Swiss precision manufacturer of injection systems for large engines, partnered with iosys to bring live production data from its proprietary, undocumented ERP onto the shopfloor. The result is a Production Control Center (PCC): ten real-time dashboards, one per production department, showing orders, deadlines and deviations as they happen. Using an AI-assisted method to reverse-engineer the ERP data model, iosys delivered a productive solution in roughly 30 engineering hours. Within three weeks of launch, DUAP was already measuring gains in quality and lead times.

Introduction

Founded in 1943 and still family-run, DUAP AG in Herzogenbuchsee (BE) builds injection systems for large engines in marine, industrial and energy applications. It is a recognised innovator in alternative fuels such as hydrogen, with deep in-house precision manufacturing, including its own hardening shop.

Like many Swiss manufacturers, DUAP had valuable production data locked inside its ERP. Out of reach of the people who needed it most: the teams on the shopfloor.

"We wanted our production teams to see, at a glance, what matters right now: orders, deadlines, deviations. That data lived in the ERP, but it never made it to where the work happens."

The Challenge

The goal: bring live production data from the ERP onto the shopfloor screens. The obstacles were substantial:

• An undocumented system. DUAP's ERP is proprietary, with no schema documentation. Roughly 5,000 database objects, using cryptic abbreviations for table and column names.
• A foreign stack. An Oracle database and a data model whose structure was not known internally.
• No guide. An expert who could have explained the data model in detail was not available.

The classic options were unattractive: a potentially slow, expensive development effort, buying in a specialist, or not attempting it at all. Instead, DUAP's management saw the potential in working with iosys: reaching the data that mattered in an agile, time-saving and highly focused way, with specific AI and data-preparation expertise, and presenting it compactly and as needed.

The Approach: AI as Investigator and Translator

iosys reconstructed the data model without further help, with a disciplined, AI-assisted method. In four steps:

1. Map the structure. List every table the system contains and sort by size. The tables holding real data stand out from the empty ones.
2. Translate the language. AI maps shopfloor concepts ("order", "scrap", "operation") onto the ERP's cryptic internal names.
3. Prove every assumption. No guess counts until a query confirms it. Hypothesis, query, evidence.
4. Anchor the knowledge in code. Once a query is proven, it moves into the application. Captured permanently in version control, instead of undocumented in people's heads.

The funnel from a sprawling database to the handful of tables that actually matter:

~5,000 objects in the ERPeverything present→~40 tables with substancesorted by size→6 core tables (shopfloor)proven by query

The rest, empty, technical or configuration tables, was irrelevant to the shopfloor.

The Data Boundary: The Customer's Data Stays Home

A question every manufacturer asks: does our production data leave the building? No. The method draws a clean line between structure and content:

Data boundaryCustomer · on-siteLive ERP dataOrders, quantities, deadlinesstays in-house→Structure←Query→ResultAI · remoteSees only the structureNames and keysnever the content

• The data stays with the customer. The AI sees only the schema, never real orders or quantities.
• Logic changes in the room. A rule can be adjusted and shown against live data in the same session, close to real time.
• The customer owns the expertise. The AI only accelerates the investigation. The truth is the customer's process plus their data. No ERP specialist required, just good questions and fast verification.

The Results

In early June 2026, DUAP officially opened its Production Control Center on the shopfloor: the same solution running on ten screens, one per department, each showing that department's own live KPIs.

The whole solution was built end to end with AI-assisted development: lean, maintainable, and fully owned by the customer.

"Just three weeks after launch, the Production Control Center has already proven its value: we're seeing measurable improvements in both quality and lead times. Even more importantly, despite some initial scepticism, our employees quickly recognised the benefits and made the system part of their daily routine. A special thank you goes to iosys for their professional support in data mining and processing. Thanks to their expertise, we were able to implement the project quickly and very efficiently."

Where the Challenges Lay

The technology was only part of the story. The harder parts were organisational:

• Daily feedback. Without a daily booking from production, there is no daily KPI. That discipline had to be established first.
• Handling scrap. How scrap is dealt with. That was change management, not just code.
• Field meanings as traps. Cryptic names, status codes and an unclear order of operations. Fortunately, that was quick to work out.

In the end, data quality is as much an organisational question as a technical one.

Conclusion

In a matter of weeks, an unreadable ERP became a productive system that ten screens now use every day. Three lessons carry beyond this project: unknown technology is no longer a major obstacle, thanks to AI. The knowledge lives permanently in the code, not in individual heads. And the method transfers to the next system. If your production data is trapped in a legacy or undocumented ERP, iosys can help you unlock it.

This project is one example of our broader work on process automation and digital shopfloor management.