AI in SMEs: High expectations, low impact
Hardly any other topic is currently being discussed as intensively as artificial intelligence. Budgets are being approved, proof-of-concepts launched and pilot projects initiated.
But the reality is sobering:
The majority of all AI initiatives never reach productive operation or generate measurable economic added value.
The problem is rarely the technology.
The problem is the structure.
The 7 most common reasons why AI projects fail
1. No clearly defined business case
Many projects start with a technological question:
"How can we use AI?"
However, the crucial question is:
"What specific problem is currently costing us time or money?"
Without measurable targets such as:
- Cost reduction
- Throughput time
- Error rate
- Conversion rate
there is no economic basis.
Success factor:
Define clear KPIs and a realistic ROI target before starting the project.
2. Inadequate data quality
AI is data-driven. If data:
- is maintained inconsistently
- is stored in silos
- has no history
- is managed manually
even the most powerful model cannot generate added value.
Success factor:
Structured data ingestion and clean feature engineering before model training.
3. False expectations of large language models
Since the breakthrough of AI providers such as
OpenAI
and tools such as
ChatGPT ,
there is often the impression that a language model can automatically solve complex business processes.
However:
- An LLM does not know your internal data without suitable architecture
- It does not replace process analysis
- It is not an ERP system
Success factor:
Use LLMs in a targeted manner – not as a universal solution.
4. Proof of concept without a scaling strategy
An AI model works in the test system.
But productive operation requires:
- Infrastructure
- Monitoring
- Security concept
- Maintenance strategy
- Scalability
Without a DevOps plan, the project remains an experiment.
Success factor:
Consider scaling from the outset.
5. Lack of internal acceptance
Technology alone does not change an organisation.
If employees:
- are not involved
- do not recognise the benefits
- lack training
- are afraid of automation
the system will be bypassed or not used.
Success factor:
Change management is an integral part of every AI project.
6. No structured project management
AI projects are interdisciplinary:
- Business
- Data engineering
- Backend
- UX
- Infrastructure
Without clear roles, prioritisation and iterative control, delays and budget overruns occur.
Success factor:
Agile project management with clear accountability.
7. Focus on technology instead of the problem
Many initiatives start with:
"Which model should we use?"
Successful projects start with:
"Which process causes the most friction losses?"
Technology is a means to an end – not the goal.
How successful AI projects are structured in medium-sized businesses
Successful companies follow a clear sequence:
- Problem definition
- Economic analysis
- Data analysis
- Architecture design
- Model implementation
- Iteration & optimisation
- Scaling
Not the other way around.
A tried-and-tested process model
Phase 1: AI readiness analysis
- Evaluation of the data structure
- Identification of relevant processes
- Assessment of economic potential
Phase 2: Business case & KPI definition
- ROI scenarios
- Investment planning
- Amortisation over time
Phase 3: Technical architecture
- API-first approach
- Clean backend structure
- Monitoring & logging
Phase 4: Iterative implementation
- MVP
- Testing
- Feedback cycles
Phase 5: Production launch & scaling
- Integration into existing systems
- Performance optimisation
- Security and compliance testing
Practical example from a medium-sized company
A company automated its quotation process.
Before:
- Manual data entry
- Long processing times
- High error rate
After implementing a structured AI solution:
- 60%+ time savings
- Significantly reduced errors
- Payback within one year
The decisive factor was not "more AI".
The decisive factor was the clean project structure.
Conclusion: AI is not a tool – it is infrastructure
AI projects do not fail because of algorithms.
They fail because of a lack of strategy.
Successful companies do not treat AI as a trend, but as a long-term infrastructure decision.
The difference is not in the budget.
It is in the approach.
Next step
If you want to check whether your company is structurally ready for AI:
- Analyse your processes
- Evaluate your data quality
- Define clear business KPIs
A thorough AI readiness analysis is often the most sensible way to get started.
