Working with AI: From Prompt to Work System

How can I work with AI?

There is often no context and no preparatory work, which results in unstructured prompts. The result is correspondingly generic and confirms the impression that AI only delivers superficial results.

The decisive difference is not in the tool, but in how it is used. Anyone who sees AI as a work system rather than as a search engine with a better formulation will achieve different results. In concrete terms, this means three things:

Build context instead of starting from scratch

Tools such as ChatGPT and Claude offer the option of storing user-defined instructions — so-called custom instructions. There, industry, role, communication style and typical tasks can be defined. Once set up, this provides answers that fit your own work context.

Work with projects and documents

Many AIs offer project functions that allow documents to be uploaded. Strategy papers, product catalogues, competitive analyses, internal guidelines: everything the model should know becomes a knowledge base. Instead of general advice, AI provides answers based on its own corporate documents. A medium-sized retail company that uploads its product catalog and target group description receives product descriptions that therefore match its own product range without first having to write an explanatory prompt.

Create prompt templates for recurring tasks

Anyone who writes a status report every week, formulates an offer or evaluates customer feedback can build and reuse a structured template for this purpose. This not only saves time when entering, but also ensures consistent quality.

Better research: From tab chaos to structured analysis

Research is one of the tasks where AI is most underestimated. Many use ChatGPT as a kind of Google replacement and expect a single, correct answer.

The strength of AI lies in the structured preparation of complex topics. Instead of a single question, a multi-stage process is worthwhile: First narrow down the topic, then ask for different perspectives and deepen them in a targeted manner. For example, anyone who wants to evaluate the use of AI in their own industry is better off to start with a request for current fields of application, then have specific case studies given and, in a third step, ask about risks and prerequisites.

Perplexity is particularly suitable for research with source references. Unlike ChatGPT, it provides answers with direct links to the sources used. This is helpful when results are shared internally or incorporated into decision templates. Google's NotebookLM goes one step further: It can be fed with your own documents and answers questions based exclusively on these sources. For teams that regularly work with extensive reports, studies, or internal documents, this can be a huge time saver.

What is often overlooked is that AI-based research does not replace one's own judgment. It speeds up the path to a well-founded assessment. Evaluating whether information is relevant, whether sources are reliable and what conclusions can be drawn from them — that remains the task of the team. If you are looking for an in-depth overview of available AI tools, check out our Overview article on AI tools and applications 2026 good orientation.

AI as a sparring partner: Better prepare decisions

One of the most valuable use cases is to use AI as a thought partner for complex issues. It helps to bring in new perspectives, uncover blind spots and to better structure your own arguments.

A specific example: A managing director is faced with the decision as to whether she should enter a new market segment. Instead of asking ChatGPT whether that's a good idea, she can ask the tool to analyze the decision from three different perspectives: from the perspective of the CFO, sales management, and a competitive analyst. The quality of answers increases significantly when the context is right: company size, industry, previous experience, available resources.

This approach is also worthwhile when preparing workshops, strategy meetings or negotiations. Anyone who plays through various scenarios in advance with AI goes into the room better prepared. This does not replace experience or industry knowledge, but it makes both more effective.

Practitioners will be showing how SMEs are specifically integrating AI tools into their everyday working life at d:u26 on March 26 & 27 in Münster. You can find an overview of all speakers here.

The right tool landscape: thinking beyond ChatGPT

Everyone knows ChatGPT. Microsoft Copilot is already part of the Office license in many companies. Google Gemini is growing. But the really exciting developments are happening beyond the big platforms. There is potential there for teams that want more than just text generation.

DeepL Write, for example, optimizes existing texts stylistically and linguistically without changing the content. For companies that regularly write offers, reports or customer communication, this is a more precise tool than general language models. Perplexity provides source-based research with references that can be checked. NotebookLM from Google turns a collection of documents into a searchable knowledge base without sharing data with third parties. The deep research functions, which are now available in ChatGPT, Gemini and Claude, create multi-page analyses based on current web sources. This is useful for market research, competition analyses or the preparation of strategy papers.

It is not the number of tools that is decisive, but the conscious selection. Anyone who uses three tools specifically gets ahead of someone who has tried ten and doesn't use any correctly. The starting point should always be the use case, not the technology. Which task takes the most time? Where do most mistakes occur? Where is capacity missing? From there, you can select the appropriate tool in a targeted manner.

Shadow AI: The growing risk of uncontrolled deployment

Precisely because AI tools are so easily accessible, a problem is arising in many companies: Employees have long been using them without consultation. Sensitive company data can thus end up in external systems, generated texts are transferred without verification, or internal information flows into models whose data processing is not controlled.

Clear framework conditions should be created in companies to control the use of AI and create a secure framework for handling their own data. The risk concerns not only data protection in the strict sense, but also trade secrets, strategic information and the quality of work results.

Good internal AI guidelines govern which tools are allowed, which information may not be entered into external systems, how AI-generated results are checked and marked, and who is the contact person for questions or problems. The effort involved is manageable, but the protection is considerable. If you want to dive deeper into the topic of data security and AI, check out our Article on data security and AI the most important basics.

How can I use AI without turning everything around?

Not every company has an innovation budget or its own data team. It doesn't need that either. What is needed is a structured start that goes beyond trying things out occasionally.

A proven approach: Define a pilot team of three to five people who systematically use AI tools in their day-to-day work over four to six weeks. Not by the way, but with a clear mandate: Which tasks are suitable? Which results are useful? Where are the limits? The results are documented and shared within the company so that other teams benefit from them.

What is often underestimated is that learning works best in exchange. Teams that share their experiences with AI tools and regularly discuss what worked, what didn't, or which prompts deliver good results, develop a sense of where AI creates real added value and where it's more likely to get in the way. This internal transfer of knowledge is more valuable than any seminar.

Let's say that a team starts with a single, clearly defined use case, such as product descriptions or offer texts, which previously cost a lot of time. After a few weeks, the time required is significantly reduced and the quality of the results remains constant. Trust in technology is growing from these initial successes, and other areas such as marketing, purchasing and customer service are following suit. The decisive factor would not be the tool itself, but the structured start with a specific problem that can be clearly measured.

From the first AI project to a company-wide introduction: at data:unplugged, teams share their experiences from everyday working with AI. Find out why d:u26 is the right event for you and your team.

Where AI shouldn't take over

As powerful as the tools are now, they also have clear limits. Strategic decisions based on experience, industry knowledge, and intuition cannot be delegated. Tasks that require empathy, such as employee interviews, difficult negotiations, sensitive customer communication, remain with people. And quality control also remains essential. AI tools can create texts, analyze data and recognize connections. But they cannot guarantee that the results are accurate and relevant in a specific business context.

In medium-sized companies, personal relationships, deep industry knowledge and the ability to act quickly and pragmatically count. AI amplifies these strengths when used correctly. It doesn't replace them. Anyone who internalizes this avoids disappointment and uses AI as what it does best: accelerate routine tasks, structure knowledge and prepare decisions, not make them.

Conclusion: Depth beats width

Most companies have made their start with AI a long time ago. The decisive factor is how well the tools are used. Anyone who builds context, works with documents, selects the right tools for the right tasks and defines clear rules of the game turns a nice tool into a real and at the same time scalable work system.

The next step doesn't have to be perfect. It just needs to happen. Set up custom instructions, create a project with the most important documents, build a prompt template for a recurring task — that's enough for initial experience and more clarity about the next steps.

Find out how other SMEs are taking this path at the data:unplugged festival 2026 on March 26 & 27 in Münster. At the Mittelstand Blazers Stage, companies share their experiences and learnings from data and AI projects. In the master classes, it becomes more specific:

• AI tools for image generation
• Agent-based process automation
• Automate complex workflows with AI agents

Working with AI affects all areas of the company — from IT to marketing to management. For successful implementation, it is important to define, involve and qualify key people. data:unplugged stands for practical transfer of knowledge, from which the entire team benefits. There are specific master classes and stage programs for a wide range of tasks. Get your ticket now!