How STADLER reduced the time spent on everyday knowledge work

The problem.

STADLER wanted employees working at computers to spend less time on repeated preparation and more time using their expertise.

What changed.

The company combined ChatGPT access, training and shared guidance with employee experimentation. Uses included documentation, translation and drafting.

As described by OpenAI and STADLER.

What was reported.

OpenAI reports 30–40% time savings on common knowledge tasks. This concerns office work around the business, rather than the throughput of its sorting equipment. Source: OpenAI

What we take from it.

In an engineering business, useful knowledge often lives across manuals, supplier documents and the experience of individual colleagues. A good first assistant should answer a narrow, recurring question from a maintained set of sources. We would start with the question employees already ask each week, identify who can approve the answer, and decide how an uncertain response should reach that person.

An answer needs enough context to be useful after it leaves the chat. A reference to a document should include its version and the section that supports the conclusion. If two documents disagree, the assistant should expose the disagreement. These details help an employee recognize whether the response applies to the equipment, customer or process they are actually working with.

Drafting needs a similar connection to evidence. We would agree the intended reader, required facts and review criteria before generating a technical document. An engineer should be able to correct a specific statement without reconstructing the whole reasoning process. Approved terminology and examples can guide tone, while responsibility for technical accuracy stays with the person who signs off the work.

A useful pilot measures the complete job: finding material, producing an answer, checking it and making corrections. Shorter drafting time is only one part of that result. We would also record outdated references, unanswered questions and the effort required to keep the source collection current. That makes it possible to decide which tasks deserve wider adoption and which still need a simpler process.

How to evaluate a similar idea.

Start with your situation and a question you can test. These are evaluation steps we would discuss before choosing an implementation.

1. 01

   Start with a repeated question

   Choose a task with a clear owner and recognizable output. Record the time employees currently spend finding and checking information.

2. 02

   Establish the source of truth

   Select approved documents, preserve versions and apply existing access permissions. Assign responsibility for keeping each source current.

3. 03

   Make answers easy to check

   Connect statements to supporting passages. Show missing information and conflicting sources before a draft reaches its final reviewer.

4. 04

   Measure the finished task

   Include checking, corrections and ongoing maintenance. Compare several real tasks before deciding whether to expand the workflow.