This is the first part of our series on AI implementation in mid-sized businesses. Here we address the strategic prerequisites: What questions should be answered before even thinking about tools? In the second part – “What defines Good Prompts for Mid-Sized Businesses” – we’ll show practical implementation examples.
Most AI implementations in mid-sized businesses fail not because of the technology, but due to inadequate preparation, unclear objectives, and the assumption that a new tool would automatically deliver better results.
AI Implementation in Mid-Sized Businesses: Why the Most Important Preparation Has Nothing to Do with Technology
The managing director of a mid-sized precision engineering company – let’s call her Ms. Müller – was highly skeptical about AI. After a failed project two years ago, she wasn’t exactly thrilled when her sales director approached her about AI. “Yet another IT revolution that’s supposed to make everything better?”
Her skepticism was justified. Most AI implementations in mid-sized businesses fail not because of the technology, but due to inadequate preparation, unclear objectives, and the assumption that a new tool would automatically deliver better results.
Before investing in AI, you should ask yourself some important questions. The first two are the most crucial – and they have nothing to do with software.
How Ready for Change Is Your Team Really?
AI doesn’t just change tools – it transforms ways of working. Roles shift. Tasks that took hours yesterday might be completed by AI in minutes tomorrow. This sounds like progress, but it can also trigger uncertainty within your team.
The decisive question is therefore not “Which AI tool suits us?” but rather: “How does our team respond to change?” This places you squarely in the realm of change management – the systematic guidance of transformation processes in organisations. Change management considers not just the technical side of an innovation, but primarily the human element: How do employees perceive the change? What fears and resistance emerge? And how can the team be involved so that those affected become active participants?
At Müller Precision Engineering, a mixed picture quickly emerged. The sales director was curious and eager to experiment. The experienced back-office administrator, however, had concerns: Would her years of expertise now be replaced by AI?
Three indicators of readiness for change:
Curiosity: Do your employees proactively ask about new methods? Or are changes primarily perceived as disruptions?
Experience: How did previous change projects go? A failed IT project like Ms. Müller’s leaves traces – both positive learning effects and justified skepticism.
Security: Do your employees feel secure enough in their roles to see change as an opportunity rather than a threat?
This is why it’s important to allow sufficient time for “mental digestion.” AI implementation is change management – not IT project execution. Don’t skip this crucial step, or you’ll implement technically successful solutions that later fail to deliver the desired benefits and full potential.
What Success Metrics Are You Setting for AI Implementation?
“We want to become more efficient” isn’t a goal – it’s a wish. And wishes are difficult to quantify.
Back to our example: For the AI implementation, Ms. Müller wanted to do things differently this time. Before any AI tool was deployed, she defined a concrete benchmark with her sales director: Creating proposal texts previously took an average of two hours. The new target: 30 minutes, with consistent quality.
Why metrics matter before implementation
Without a baseline, there’s no comparison. Many companies can’t say after a year of AI usage whether the investment was truly worthwhile – because they never measured what things were like before.
Clear benchmarks also create acceptance. When your team sees that a process is measurably improving, resistance to change decreases.
Practical questions for your measurement might include: How much time do your employees currently spend on the process you want to improve? What’s the error rate? How satisfied are internal or external customers with the outcome?
And an often-forgotten question: How many processes do you actually want to integrate? At Müller Precision Engineering, the answer was: One. Only when that works does the next one follow. This aligns with our “Evolution, Not Revolution” approach.
What Exactly Do You Want to Achieve with AI Integration – and Where’s the Bottleneck?
“We want to use AI” isn’t a strategy. The question must be: What specific problem do you want to solve? Better yet: What does the process look like that needs optimising? From an AI perspective, it’s processes that are interesting – and they often need to be defined first.
At Müller Precision Engineering, the bottleneck was clearly identifiable. The sales director spent eight hours per week turning technical specifications into comprehensible proposal texts. Eight hours of highly qualified working time on a task that was important but not value-adding.
The difference between automation and elimination is that AI automates tasks – it doesn’t eliminate them. The sales director still writes the proposals. But instead of starting from scratch, he begins with a draft that he reviews and refines. The expertise remains with the human; the routine work is handled by the machine.
How to identify your bottleneck
Ask yourself the following questions: Where do qualified employees spend time on tasks below their skill level? Where do delays occur because someone is waiting for input? Where do similar activities repeat with slight variations? What do the processes look like, and are they defined?
Which Processes Are Actually Suitable for AI?
Not every process benefits from AI. And not every process that could benefit is ready for it.
Three prerequisites for AI-suitable processes:
Repeatability: Processes that recur in the same or similar form are better suited than one-off special cases.
Documentation: If a process exists only in an employee’s head, no AI can improve it. At Müller Precision Engineering, the sales director first had to document how he actually proceeds, what information he needs, and which phrases resonate well with customers.
Digital availability: Can the relevant information be processed on a computer? A handwritten notebook isn’t AI input.
The honest assessment in our practical example went like this: Ms. Müller discovered that of ten processes her team deemed “AI-suitable,” only three actually met the prerequisites. This clarity helped enormously – better to properly improve one process than half-heartedly tackle ten.
Do Your Teams Work Digitally – and If So, How?
This is a pragmatic question that’s often overlooked: Does the work you want to improve actually take place on a computer, and is the data available for AI?
For a machine builder like Müller Precision Engineering, this wasn’t a trivial question. In production, people work with their hands and machines, not with laptops. Sales, however, works digitally – emails, spreadsheets, word processing.
What this means for your AI strategy:
Start where digital working methods are already established. This lowers the entry barrier and produces visible results faster. Production can follow later, once the organisation has learned how AI implementation works.
AI-Readiness Check
Three Prerequisites for AI-Suitable Processes
Repeatability
Documentation
Digital Availability
“We want to use AI” isn’t a strategy. The question must be: What specific problem do you want to solve? Better yet: What does the process look like that needs to be optimised?
What Technical Level Do You Actually Need?
Many companies overestimate the technical effort required. AI implementation doesn’t automatically mean software development.
The three levels of AI implementation:
Level One: Pure Prompting. You use existing AI tools like ChatGPT or Claude and learn to formulate good requests. No installation required, no programming – ideally with a business licence. This is sufficient for an estimated 90 percent of use cases.
Level Two: No-Code Workflows. You connect different tools so that information flows automatically. For example: An email arrives, is automatically summarised, and the summary appears in your project management tool.
Level Three: Custom Integrations. This is where actual programming happens – custom interfaces and custom applications. This is complex, expensive, and unnecessary in most cases.
The decision at Müller Precision Engineering went like this: The sales director started at Level One. He learned to write structured prompts that generate comprehensible proposal texts from technical data sheets. After three months of successful use, they considered whether Level Two would make sense. The answer: not yet. Level One solved the problem adequately.
What Tools Are You Already Using – and Do You Want to Keep Them?
AI must fit into your existing work environment. If your team works with Microsoft 365, a tool that integrates there makes sense. If you use Google Workspace, different considerations apply.
The integration question in our practical example was resolved as follows: Ms. Müller deliberately chose not the “best” AI solution, but the “most suitable” one. The sales director could stay in his familiar environment and didn’t have to switch between different applications. Less friction means higher acceptance.
How Do You Evaluate the Actual Value of AI Implementation?
Not every AI application delivers the same added value. And beyond a certain point, there are diminishing returns.
The ROI reality at Müller Precision Engineering: The sales director saved six hours per week. At an internal hourly rate of €80, this equates to savings of approximately €25,000 per year. The investment in training and tools was under €3,000.
But: This calculation only worked because the process was clearly defined beforehand, the time savings were measurable, and the sales director actually used the time gained for value-adding activities.
After the first successful process, the sales director immediately wanted to optimise three more. Ms. Müller applied the brakes: “Let’s first see if the benefit holds. And whether the team is ready for more change.” Six months later, the second process followed. Not because it couldn’t have been done faster technically, but because sustainable change in the spirit of change management takes time.
Conclusion: The Right Sequence Makes the Difference
Ms. Müller now has a company that uses AI. Not everywhere, not for everything – but where it demonstrably adds value. The difference from failed AI projects wasn’t in the technology, but in the targeted preparation.
Additionally, we can draw on our AI transformation journey to highlight these five phases of AI integration:
AI integration follows recognisable development patterns. Companies typically progress through five phases: from unplanned experimentation through disappointing tool purchases to strategic AI excellence. Each phase has characteristic challenges, measurable ROI improvements, and specific change management requirements. Learn more here.
The AI Transformation Journey
5 phases from chaos to strategic AI excellence
"Wild West"
"Copilot Purchased, Problem Unsolved"
"AI as Personal Sparring Partner"
"Multiplayer AI Teams"
"Human-led, Agent-operated"
Let’s Start a Conversation
Are you facing the challenge of strategically embedding AI in your organisation? Let’s talk about your individual transformation path.
Our promise: We know every step of this journey from first-hand experience – including all the highs and lows.
You can schedule a complimentary initial consultation at a time that suits you on our contact page.
Want to dive deeper into the strategic aspects of AI transformation? Read our article on change management as an underestimated success factor.
AI Implementation: Crafting Effective Prompts for SMEs and Family Businesses
[…] the strategic groundwork (more on that in the first part of this series), many businesses face a familiar challenge: employees have access to AI systems, but the quality of […]