ChatGPT can write a project report for you in three minutes that used to take two hours. At the same time, it can convincingly explain why your company was supposedly founded in 1987—even though you only started in 2010.
This discrepancy between impressive strengths and glaring weaknesses makes large language models a challenge for managing directors. Where does it make sense to use them? What expectations are realistic? And how do you implement AI successfully without falling for the hype?
As a decision maker, you don’t need academic essays about transformer architectures. You need clear answers to the question: What do ChatGPT & Co. tangibly deliver for my business?
What Large Language Models Are Actually Capable of Today
Large language models like GPT-4, Claude, or Gemini have made remarkable progress over the past two years. But what does that actually mean for your daily work?
Text Processing and Document Creation
The main strength of today’s AI models is text processing. ChatGPT can turn your key points into structured reports, write professional emails, or translate complex topics into clear language.
A practical example: You give the system the basic facts of a client project, and within minutes, it delivers a structured offer. Of course, you still need to check and adapt the numbers and details. But the basic structure is already there.
The models truly excel at text optimization. They can turn a clumsy email into a professional message, or summarize your notes into a presentation-ready brief.
But beware: copy-paste prompts are useless. A good prompt is like a precise specification document: the more detailed your instructions, the better the result.
Data Analysis and Summarization
Modern AI systems can swiftly comb through large volumes of data and organize them. Upload a 50-page market analysis, and the system will return the key takeaways in five bullet points.
Anthropic’s Claude, for example, can process PDFs up to 200,000 characters—that’s about 80-100 pages of text. For most business documents, that’s more than enough.
The systems recognize patterns in your data, identify trends, and can compare different documents. If you have monthly sales reports from different regions, AI can quickly spot where the biggest deviations are.
However, AI only interprets what you provide. Domain-specific knowledge or industry expertise needs to be derived from your text.
Automation of Routine Tasks
Large language models are excellent for repetitive tasks with clear rules: email categorization, extracting appointment suggestions from messages, or generating standard answers to common customer inquiries.
A mechanical engineering company in Bavaria, for example, uses ChatGPT to automatically turn unstructured client requests into structured briefings for its sales team. This saves about 15 minutes of prep time per inquiry.
The models also deliver strong results when it comes to translation. For common languages, they now reach professional quality—at least for standard texts without highly specialized terminology.
The Real Limits of Current AI Models
Hype doesn’t pay salaries—realism does. That’s why it’s important to be honest about the current limitations of ChatGPT & Co.
Hallucinations and Factual Errors
The biggest problem of current large language models is so-called hallucinations. The systems make up facts that sound plausible but are actually false.
Take this real-world example: a company had ChatGPT create a market analysis, receiving detailed figures on market shares and revenues. The problem? Half the cited studies didn’t exist, and the numbers were fabricated.
Even the best models still make factual errors more often than not. With more complex topics, this error rate increases further.
The basic rule: never blindly trust AI-generated facts. Every important piece of information must be checked.
Context and Currency Limitations
Even the latest models have a limited context window. GPT-4 can process about 128,000 characters at a time—it sounds like a lot, but it’s not enough for extensive manuals or large datasets.
On top of that, most models have a fixed data cutoff for training. GPT-4, for example, was trained on data up to April 2023. It knows nothing about the most recent developments, new laws, or market shifts.
This is especially problematic in fast-changing areas like compliance or technology standards.
One solution is retrieval-augmented generation (RAG) systems, which include up-to-date information from your own data sources. But even that requires the right technical implementation.
Limits to Handling Complex Decisions
ChatGPT can support decision making by creating pros and cons lists or simulating various scenarios. But these systems can’t—and shouldn’t—make your final strategic choices.
Especially for high-stakes decisions, those involving incomplete information or ethical dimensions, these models hit their limits. They have no true understanding of corporate policy, risk assessment, or long-term consequences.
An IT director told us: “ChatGPT gave me a perfect breakdown of why a cloud migration makes sense. But it couldn’t assess whether our 15-year-old ERP provider could actually handle the technical transition.”
Concrete Use Cases for Medium-Sized Businesses
Enough theory. Where can you actually use ChatGPT & Co. in your company?
Quote Generation and Requirement Specifications
This is one of the biggest opportunities for mid-sized companies. Project managers often spend hours creating similar quotes and requirements documents. AI can drastically reduce this time.
Here’s how it works: you provide the system with your standard building blocks, project details, and specific client requirements. The system produces a first draft, which your experts then revise and finalize.
An automation company in Baden-Württemberg reports a 60% time saving for offer generation. Important: the technical review and adjustments remain with your engineers.
Technical documentation can also be produced more efficiently. The system can generate easy-to-understand user manuals or maintenance instructions from your product specifications.
Customer Service and Internal Communication
AI-powered chatbots can answer standard customer inquiries around the clock. But beware of overpromising: the systems are still overwhelmed by complex technical questions or individual problem-solving.
A more realistic use: handling FAQs, scheduling appointments, or initially categorizing incoming requests. For example, an industrial services provider uses ChatGPT to automatically extract service appointments from unstructured emails.
Internally, the systems can support email communication—summarizing long threads, extracting key information, or turning meetings into structured minutes.
One practical example: after a two-hour project meeting, a project manager uploads the recording and automatically receives a to-do list with responsibilities and deadlines.
Knowledge Management and Training Materials
Many mid-sized companies have their knowledge scattered across different systems: ERP, CRM, file servers, individual notes. AI can help make this knowledge accessible.
With RAG systems, you can set up an in-house “knowledge chatbot.” Employees ask questions and receive answers based on your own documents, manuals, and process descriptions.
The models are also strong in creating training materials: they can turn technical manuals into accessible introductions for new hires, or translate complex processes into simple step-by-step instructions.
A mechanical engineering firm uses ChatGPT to turn technical maintenance manuals into accessible video scripts for its service team. Time saved: about 70% compared to manual creation.
What Managing Directors Should Watch Out for When Implementing
The technology is available. The question is: how do you successfully deploy it in your business?
Data Protection and Compliance Requirements
This is the issue that keeps many managing directors awake at night—with good reason. The GDPR applies to AI systems too, and violations carry serious penalties.
Basically, you’re choosing between cloud-based services (ChatGPT, Claude) and local solutions. With cloud services, your data leaves your company—that’s not automatically GDPR-compliant.
Since 2024, OpenAI has offered EU-hosted versions of ChatGPT governed by European data protection laws. Anthropic and Google have similar offerings. Still, you should never enter sensitive customer data or trade secrets into public AI systems.
For critical use cases, local models are an option. Companies like Ollama or Hugging Face offer solutions that run entirely within your own IT infrastructure. The effort is higher, but you keep full control over your data.
Our advice: start with non-critical use cases and gradually build up GDPR-compliant solutions.
Employee Enablement and Acceptance
The best AI technology is useless if your employees don’t use it or use it incorrectly. Change management is key.
Many employees fear that AI will make their jobs obsolete. That’s understandable, but usually unfounded. AI automates tasks, not jobs—your employees can focus on more valuable work.
Successful companies invest in training and hands-on workshops. Show specific use cases and let teams experiment for themselves. One day of training with real-world examples is worth more than ten PowerPoint presentations.
Importantly: establish clear guidelines for AI usage. What is allowed, what isn’t? How do you handle generated content? Who is responsible for quality control?
A mid-sized machinery manufacturer has appointed “AI Champions” in every team—these individuals receive intensive training and then help their colleagues get started.
Measuring ROI and Tracking Success
How do you track the success of your AI initiatives? Many companies invest in the technology but never measure whether it actually pays off.
Start with simple metrics: time saved on specific tasks, reduction in processing times, quality improvement. An equipment manufacturer measures how much time their engineers need to create technical documentation—before and after rolling out AI.
But be wary of false savings. Time savings are only valuable if that saved time is put to better use. If your employees end up sitting idle, you gain nothing.
Soft factors are harder to measure but just as valuable: employee satisfaction, reduction in repetitive work, improved output quality.
Our tip: launch pilot projects in clearly defined areas. Measure closely, then scale to other parts of the business.
Practical Recommendations for Getting Started
That’s enough reading—what should you actually do next?
Step 1: Identify Low-Risk Use Cases
Start where mistakes aren’t critical—drafting emails, internal meeting notes, initial document drafts. Gain experience before automating business-critical processes.
Step 2: Train Your Employees
Invest in prompt engineering training—a well-written prompt makes the difference between “I could have done this myself” and “Wow, this saved me two hours.” Many companies underestimate this point.
Step 3: Establish Clear Guidelines
What’s allowed in AI systems? Who checks outputs? How do you label AI-generated content? These rules must be set before the first employees get started.
Step 4: Measure and Adapt
Document time spent before and after AI implementation. Gather feedback from your teams. What works, what doesn’t? Adjust your approach accordingly.
Step 5: Scale Up Gradually
Only when your first use cases are running smoothly should you expand to other areas. Rushed full-scale implementations usually lead to frustration and resistance.
Don’t forget: AI is a tool, not a cure-all. The most successful companies are those that approach the technology realistically and systematically integrate it into their processes.
If you need support—whether training your teams or technical implementation—get in touch. At Brixon, we help medium-sized businesses successfully and securely integrate AI into their business processes.
Frequently Asked Questions
Is ChatGPT GDPR-compliant?
Since 2024, OpenAI has offered EU-hosted versions of ChatGPT under European data protection law. Still, you should never enter personal data or trade secrets into cloud-based AI systems. For critical applications, local models are the safer choice.
How can I spot AI hallucinations in the output?
Double-check all factual statements, especially numbers, dates, and references. Be skeptical about overly specific details or when the system mentions studies or statistics without verifiable sources. Set up a four-eyes principle for all important AI-generated content.
What cost savings are realistically achievable?
For text tasks like quote generation or documentation, time savings of 40–70% are realistic. Important: the saved time must be put to good use. Saving time without an increase in productivity brings no ROI. Start with measurable pilot projects.
Do I need my own IT infrastructure for AI?
For starters, cloud services like ChatGPT or Claude are sufficient. For more data-sensitive applications or specific requirements, local models can be useful. These require adequate hardware and IT expertise. Many medium-sized businesses start with cloud solutions and gradually build their own capacity.
How long does it take to introduce AI into a business?
Simple use cases can be implemented within weeks. Company-wide AI strategies with training, guidelines, and technical integration take 6–12 months. Crucial is a step-by-step approach: pilot first, then scale.
Will AI replace my employees?
AI automates tasks, not jobs. Your employees will be freed from repetitive work and able to focus on strategic, creative, or advisory activities. Effective change management is vital to dispel fears and communicate the benefits.
What does it cost to start with enterprise AI?
Cloud services start at €20–50 per user per month. Add training costs and possible adjustments to your existing systems. For a systematic start with workshops and pilot projects, expect to budget €10,000–25,000. The ROI usually materializes within 6–12 months.
