Many medium-sized companies are currently launching AI initiatives with high expectations. However, after initial pilot projects, the picture often looks quite different: the technology works, but there are no measurable benefits. The reasons for this rarely lie in the AI itself, but rather in typical structural patterns that run through many implementation projects.
Mistake 1: Unclear goals – “Gather experience first”
Quite a few AI initiatives start with a reasonable idea: you want to explore the topic without immediately risking a large investment. So a pilot project begins. A team tests a system, gathers impressions, and reports back to management.
What is missing is a simple question: What specific problem should be solved as a result?
Often, there is only a general goal. Increase efficiency. Save time. Become more modern. That sounds plausible, but it doesn't lead to any decisions. Because no matter how the project goes, it's hard to evaluate it in the end. The participants have learned something, but the company hasn't changed anything.
AI doesn't generate benefits through use alone. It generates benefits when it speeds up a decision, eliminates a work step, or reduces a source of error. Without a clearly defined business problem, there is therefore no economic effect. The pilot project remains an experiment.
That's why successful implementations seem surprisingly unspectacular. They don't start with the question of which AI to use, but with another: Which task regularly costs us time, money, or nerves today and would be noticeably easier if it were done differently tomorrow?
Mistake 2: Proliferation of tools instead of strategy
In many companies, AI is not created by a single decision, but by many small ones. An employee tests ChatGPT, IT activates Copilot, marketing tries out an image tool, sales discovers an automation solution. Each of these decisions makes sense on its own.
However, the real problem is not the number of tools, but the lack of commitment. AI is used when someone is interested or has time. This means it remains a personal tool and does not become part of the way of working.
The effect only occurs when AI is embedded in processes. An offer is automatically prepared, a request is pre-structured, a malfunction is pre-analyzed. The benefit then no longer depends on individual people, but on the process. This is exactly where scaling begins.
Companies therefore rarely need to ask which tool is the best. Another question is crucial: At what point in the process must a result be reliably produced?
Mistake 3: Data is available but not usable
Almost every medium-sized company has large amounts of data. Quotes, service reports, complaints, emails, drawings, and logs have accumulated over the years. So there is no shortage of data.
Nevertheless, it is of little use. The reason lies in the difference between existing data and usable knowledge. Information is scattered across folders, network drives, and mailboxes, names differ, and multiple versions exist. Crucial information is found in free text fields or attachments. People with experience can find their way around this, but systems can hardly.
AI can only work reliably if connections are recognizable. Which information belongs to which process? Which source is current? Which information is binding? If these questions remain unanswered, the results appear arbitrary. Employees check again themselves and the application loses its significance.
That's why effective AI deployment doesn't start with a model, but with structure: Could a new employee work with your data tomorrow without having to ask anyone first?
Mistake 4: Processes remain unchanged
AI rarely replaces individual tasks. It changes how work is organized. Steps are shifted, decisions are made earlier, and preparation happens automatically. If the process remains the same, the result remains the same. The application may save minutes, but it does not change capacity.
That is why many initiatives fizzle out despite functioning technology. The employees did not work incorrectly. They simply continued to work as before. AI only becomes effective when processes are adapted. A request is no longer read in its entirety, but is handed over in a pre-structured form. A quote is created from existing information instead of a blank document. A malfunction is first analyzed and only then forwarded. This not only increases speed, but also reduces workload: Which tasks do employees repeat every day, even though the information has long been available?
Mistake 5: Employees are involved too late
In many companies, AI is first decided upon by management or IT and then presented. This works to a limited extent for office areas, but usually hardly at all for production, service, and maintenance.
Those who work with machines, systems, or customer problems on a daily basis were not involved in the selection process. As a result, the application often does not fit the actual workflow. A service technician needs information in seconds, not a menu. A machine operator cannot make long entries during a malfunction. So work continues as before.
Open resistance rarely arises. Instead, AI is circumvented. Problems are solved based on experience, colleagues are called, and notes remain handwritten. At the same time, initial successes are emerging in other areas. As a result, AI appears to be a topic for individual departments rather than a common work tool.
Added to this is uncertainty: Can I rely on an analysis? Who is responsible for errors? As long as these questions remain unanswered, the use of AI will be hesitant or non-existent.
Only when the employees concerned are involved at an early stage can meaningful applications be developed, such as a pre-structured service history, faster fault diagnosis, or comprehensible shift reports. So the question would be: Does AI arise where decisions are made or where the work happens?
Why is AI not reaching everyday working life in small and medium-sized businesses?
The reason often lies in the transition from testing to commitment. In the pilot project, a small team works with time and attention. In everyday life, timing, responsibilities, and liability are what count. A solution must become a reliable part of the process there, not just be used occasionally.
This is where AI often gets stuck. Its use is permitted, but not required. Results are not automatically incorporated into decisions. Offers continue to be created without preparatory work, disruptions are passed on without analysis, and plans are made without forecasts. The process continues to function even without AI.
This does not result in any real productivity gains. Individuals work faster, but the company does not. Progress depends on committed individuals rather than the process. However, scaling only begins when a rule changes: a work step is only considered complete when the result has been used. Only then does a functioning experiment become operational practice. AI is then no longer a helpful tool for individual employees, but the way the company works.
Conclusion
As long as processes remain the same, the effect of AI in small and medium-sized businesses will remain limited. A pilot project can show that something is possible. Economic benefits only arise when work is distributed differently, decisions are prepared earlier, and results are incorporated into processes in a binding manner. Then AI doesn't just save minutes, it changes capacity.
The decisive step is therefore not another test or a better tool. The key is to determine where AI will become part of the work and which tasks will be based on it in the future. Only then will scaling, trust, and thus ROI emerge.
AI is not an additional capability of a company. It becomes part of its organization. Companies that clarify this early on benefit quickly. The others continue to gain experience.
FAQ
Is artificial intelligence also worthwhile for small businesses?
Yes. The benefits depend less on the size of the company than on the repeatability of tasks. Wherever similar processes occur on a daily basis, AI can save time and reduce errors. Small businesses often benefit particularly quickly because coordination is shorter and changes can be implemented more quickly.
Why do many AI projects fail to deliver ROI?
Because they are introduced as a test or additional tool. As long as employees can continue to perform tasks entirely on their own, the capacity of the company does not change. Economic benefits only arise when results are incorporated into processes and decisions are based on them.
Which areas in medium-sized businesses are particularly suitable for AI?
Typical entry points are quote preparation in sales, inquiry pre-structuring in service, documentation evaluation, demand forecasting in purchasing, and error analysis in production and maintenance. Areas with recurring information and clear decisions are particularly suitable.
Is AI an IT project?
Only partially. The technical introduction is usually the smaller step. The bigger step is to determine where results will be used in a binding manner in the future and who will make decisions based on them. That is why AI is primarily an organizational and management issue.
Do employees need to be able to program?
No. Most applications can be used without programming knowledge. It is more important to be clear about what AI is to be used for and how results are to be checked. Expertise in your own process is more important than detailed technical knowledge.
How do you get started with AI in your company?
Not with a tool, but with a task. A concrete, recurring process that takes time or generates errors is best suited. Only when it is clear what result needs to be improved is a suitable application selected and integrated into the process.
Posted by:
Jens is the co-founder of bakedwith, a boutique consultancy firm specialising in smart automation and AI. He helps medium-sized companies and corporations to optimise processes, reduce manual work and achieve growth through efficient workflows. Previously, he was Growth Lead at OMR and worked as a freelancer, helping numerous companies optimise their CRM and automation systems. Jens is passionate about combining growth and efficiency to enable teams to focus on what matters most.
