Skip to content
DataFlowFacts
Hardware check Configure your own AIConfigurator
DE

Case studies

The knowledge is there.
The question is how quickly the next person finds it.

Two companies that work very differently and had the same problem: knowledge that exists, is valuable and is scattered. Two stories about how their own AI on their own premises changes access to it. With labelled figures and no invented quotes.

PrintsPaul · Mechanical engineering

25 years of machine knowledge. Instantly findable.

Machine documents, maintenance information, spare parts knowledge and the experience of long-serving employees: new colleagues ask their own AI and get the answer with its source.

Spieletechnik.tv · Broadcast technology

Production knowledge stays production knowledge, even when staff change.

Workflows, software, interfaces and configurations from past productions become an internal technical knowledge assistant.

Case study 01 · PrintsPaul

25 years of machine knowledge.
Instantly findable.

PrintsPaul works with printing, converting and inspection solutions and numerous custom machines. Every machine comes with documentation, and every project with experience that is recorded somewhere. Or held in someone's head.

Mechanical engineering · Printing & converting Use case: Company knowledge Your own AI on your premises

Before: “Who knows that?”

A company with knowledge that has grown over decades. It sits in machine documents, operating manuals, maintenance information, technical documentation, spare parts information, old project files and internal documents. And in the experience of employees who have been with the company for many years. Anyone who wanted to know something first asked for the person who might know it.

The problem

This knowledge exists. But it is not automatically available. An experienced employee remembers that there was a similar case once before. A new employee does not know that. They search on drives with folder structures that have grown over the years, in PDF collections for individual machines and in e-mails from old projects. And if they find nothing, they interrupt a colleague who is busy with something else.

A typical technical question can easily cost 30 minutes, often more. With 25 years of machine knowledge, that becomes a daily problem.

The change: “Ask your own AI.”

The idea was not to change the knowledge but the access to it. The company's own AI system became the internal knowledge interface: it reads the existing documents where they are stored and answers questions with a reference to the document the answer comes from. A new employee does not first have to know which colleague has been working on this machine for 15 years. They ask the AI, read on in the document and only ask a colleague when it is really necessary.

The AI does not change the knowledge. It changes who can access it, and how quickly.

Today: how employees ask

The answer comes with a reference to the source, so it remains traceable.

  • “How does this machine work?”
  • “What maintenance is scheduled for it?”
  • “How was this fault solved in a similar project?”
  • “Which spare parts belong to it?”
  • “Where do I find the technical documentation?”

Result

Answers with a source instead of searching across drives. Fewer queries to individual knowledge holders. An onboarding path for new employees that does not depend on who happens to be in the building. The orders of magnitude we work with in projects like this are shown here as an example scenario. The actual values are measured together with PrintsPaul in live operation.

30 → 7 minTypical technical question: searching and asking around before, an answer with source afterwards. Example scenario
6 → 4 weeksExample onboarding until a new employee answers machine questions themselves
10 → 4Recurring queries to experienced colleagues per week. Example scenario
470 hrsAnnual effort for searching and queries with 20 employees, example calculation below
€25,850Example productivity value per year; conservatively at 50 %: 235 hours, €12,925
With sourceEvery answer points to the document it comes from

Example scenario with assumed values, not a customer measurement. Actual values are measured in the customer project and added here once they are reliable.

The bigger effect

What used to sit in one colleague's head becomes company knowledge. Individual experience becomes a body of knowledge the whole company can access. That applies to the new employee in their first week just as much as to the day an experienced colleague retires. And the experienced staff are interrupted less often, because the standard questions are answered before they are asked. The value lies not only in minutes saved: in faster onboarding, in knowledge that stays in the company, and in an operation that depends less on who happens to be in the building.

Example calculation

What does it cost not to find knowledge? A simple calculation with assumed values shows the order of magnitude: first today's effort, then a deliberately conservative assumption of how much of it is saved.

Example calculation

20 employees, 30 minutes of searching and queries per week

20 employees × 30 minutes per week
10 hours per week
10 hours × 47 working weeks
470 hours per year
470 hours × €55 internal full cost
€25,850 per year
Conservative assumption: only half of it is saved
235 hours, €12,925 per year

Example calculation with assumed values, not a customer measurement. Enter your own values for users, time saved and hourly rate in the calculator.

Calculate your own case

The local setup

The company's own AI infrastructure in-house, connected to the existing repositories holding machine documentation, maintenance records and project folders. Roles and access areas follow the existing permission structure: anyone who is not allowed to open a folder gets no answer from it either. Knowledge processing runs on the hardware inside the company. There is no cloud account that documents are uploaded to, and no AI service that processes them.

  • Use case: company knowledge with source references, onboarding
  • Sources: machine documentation, maintenance and spare parts records, project folders
  • Permissions: according to the existing folder and role structure
  • Scope: hardware, setup, connection, permissions, joint testing and support from a single source

Next steps

Connect further knowledge sources, refine roles, collect metrics. And gather the questions employees actually ask: they show best where the next knowledge lies that has not yet been made accessible.

Case study 02 · Spieletechnik.tv

Production knowledge stays production knowledge, even when staff change.

Spieletechnik.tv develops technology for game and quiz shows: real-time graphics, voting, game logic, statistics, editorial systems, social media integration, control systems, timing, much of it as in-house software development. A company made of knowledge and technology.

Broadcast technology · Software development Use case: Technical knowledge assistant Your own AI on your premises

Before: reconstructing instead of looking up

In complex productions, knowledge does not consist of documents alone. It sits in workflows, software, interfaces, configurations, control logic, technical particularities, custom solutions and lessons learned. Every show is a technical setup, and every setup a solution that might be needed again next time. Configurations were kept in project folders, particularities in notes, interface descriptions in the code or in the memory of the person who built them.

The problem: time pressure

Before a production, time is short. That is exactly when the questions come up: how did we solve this last time? Which component was that? Who built it? Anyone who was on the last production knows how the voting was connected. Anyone who is new or looked after a different show reconstructs what had already been solved. That costs hours in preparation and makes the company dependent on the right people being available at that moment.

If production knowledge must not be lost, it is not enough for it to be written down somewhere. It has to be findable under time pressure.

The change: an internal technical knowledge assistant

Not a “ChatGPT replacement” but an assistant for the company's own production knowledge. The AI does not replace the technicians and developers. It helps them access existing knowledge faster: setups from past productions, interface descriptions, configurations, control logic. A new technician does not first have to know who built the setup back then. Every answer points to its source, so it remains traceable and the technician can read on from there.

Today: how technicians ask

The answer is linked to sources from the internal knowledge base, so the technician can read on from there.

  • “How did we solve this in the last setup?”
  • “Which component was used?”
  • “Which interface does this system need?”
  • “Where is the documentation for this?”
  • “What was special about the last production?”

Result

Faster reuse of past solutions, less dependence on individual knowledge holders, an entry point for new team members through the company's own knowledge base. The orders of magnitude we work with are shown here as an example scenario. The actual values are collected together with Spieletechnik.tv.

120 → 45 minReconstructing a previous setup before a production. Example scenario
75 minTime gained per case when a setup is looked up instead of reconstructed
20 casesAssumed cases per year in which a previous setup is needed
25 hrsTime gained per year, equivalent to €1,625 at €65 internal full cost per hour
In-houseSource code, setups and configurations are processed on the company's own hardware
With sourceEvery answer points to a project folder, documentation or code

Example scenario with assumed values, not a customer measurement. Actual values are measured in the customer project and added here once they are reliable.

The bigger effect

Production knowledge stays in the company, even when staff change. Before a production, nothing is reconstructed any more; it is looked up. Anyone new to the team finds their way in through the questions they have anyway, and ends up at the documents they should read. The knowledge of experienced staff becomes reusable without them having to tell it all over again each time. The economic core here is less the number of hours than the resilience: a production no longer depends on whether the one person who built the setup back then can be reached.

Example calculation

Deliberately calculated small, only with the time that can be measured. We do not include lost revenue or missed productions because we cannot substantiate them.

Example calculation

Reconstructing a previous setup: 120 instead of 45 minutes

120 minutes to reconstruct, 45 minutes to look up
75 minutes gained per case
75 minutes × 20 cases per year
25 hours per year
25 hours × €65 internal full cost
€1,625 per year
Not included
Knowledge retention, onboarding, independence from individuals

Example calculation with assumed values, not a customer measurement. Enter your own values in the calculator.

Calculate your own case

The local setup

The company's own AI system in-house, connected to project repositories, technical documentation and configuration records. Software, configurations and technical documentation are the company's intellectual property. Knowledge processing takes place within its own infrastructure: no account with an AI provider, no source code or setups uploaded to a service. We do not claim that absolutely no data ever leaves a company; we describe what the system does: it processes on the hardware in-house.

  • Use case: company knowledge and coding, meaning understanding documentation, configurations and code
  • Sources: project folders from past productions, technical documentation, interface descriptions
  • Permissions: by project and role
  • Scope: hardware, setup, connection, permissions, joint testing and support from a single source

Next steps

Connect further project repositories, structure interface descriptions, evaluate the questions from everyday production work and derive from them which knowledge should be documented next, so it is not tied to one person again.

The economic core

Where the value lies.
Not only in minutes saved.

Both example calculations count only time, because time can be measured. The larger part of the benefit appears in no hourly calculation. Eight effects we see in projects.

TimeAn answer with source in minutes instead of searching across drives and asking around.
ProductivityExperienced colleagues are interrupted less often and work on what they are needed for.
OnboardingNew employees ask the AI instead of the colleague who has known it for 15 years.
Knowledge retentionExperience stays in the company, even when someone retires or moves on.
ResilienceOperations depend less on whether the one person happens to be reachable.
GrowthMore users, more sources, new capabilities: on the running system, without starting over.
ContinuityHardware and knowledge base remain, even when the AI model changes.
IndependenceNo mandatory subscription, no provider account: you decide on costs and direction.

For context

Real customers, labelled figures, no invented quotes.

Both companies have approved being named as references. The stories describe the starting point, the change and the setup editorially. Figures are labelled as example scenarios and are measured in the customer project. We publish verbatim statements only with the approval of the respective speaker; the approved testimonials from Paul Arndt and Igor Milan can be found on the start page. We take the same approach to security statements: explaining instead of promising.

How much company knowledge is inside your company?

In folders, in projects, in heads. The configurator shows in a few minutes what your own AI costs for your case. Or you discuss your use case directly with our engineering team.

Start configurator Discuss your use case