TL;DR: McKinsey reports over 70% of AI projects fail. The three most common causes: technology before need, scope too large, and failing to involve the people doing the work. The countermodel: one project, one goal, maximum 3 months, one responsible person.
According to a McKinsey study, over 70% of all AI projects in companies fail. For SMEs, the rate is likely even higher. Not because the technology doesn’t work. The approach is wrong.
From my experience as a Digital Leader and AI consultant, I see the same three patterns over and over.
Trap 1: The Technology-First Problem
“We need AI!” “What for?” “We’ll figure that out.”
Sounds absurd, but it happens constantly. A CEO reads about ChatGPT, attends a trade show, hears a keynote, and comes back with the mandate: “We need to do something with AI.”
What happens: An expensive tool is purchased, a pilot project without a clear goal is launched. After 6 months, nobody has seen measurable benefit. The budget is gone, and so is the motivation.
A typical example: A Zurich-based services company with 80 employees purchased an AI platform for CHF 45,000 per year. After 8 months, they were only using the platform for basic text formatting. The same functionality could have been delivered by a ChatGPT Plus subscription at CHF 240 per year.
How to do better: Never start with the technology. Start with the problem:
- Where is your company losing time or money?
- Which processes are repetitive and error-prone?
- Where are employees making decisions based on incomplete data?
Only when you have a concrete problem should you check whether AI is the right solution. Sometimes a simple Excel macro is better than an AI system.
Trap 2: The Big-Bang Approach
“We’ll transform everything at once.”
The SME hires a consultant who designs a comprehensive AI strategy: chatbot for customer service, predictive analytics for procurement, automated quality control, AI-powered marketing. All at once.
What happens: The team is overwhelmed. Systems aren’t integrated. No single project gets enough attention. After a year, there are many started and no completed projects.
Here is the typical escalation curve of the big-bang approach:
Month 1-2: High energy, many ideas, kick-off meetings
Month 3-4: First difficulties, delays, scope creep
Month 5-6: Team frustration, budget pressure
Month 7-8: "We need more time / more budget"
Month 9-12: Project put on ice or significantly reduced
Result: High costs, barely measurable value, AI skepticism in the organizationHow to do better: The One-Project Approach:
- Choose one project with a clear, measurable goal
- Define a timeline of maximum 3 months
- Assign one responsible person (not a committee)
- Measure success with one concrete metric (time saved, error rate, cost)
Only when this project succeeds do you start the next one. Each completed project builds competence and trust.
Trap 3: Forgetting the People
“The AI does this for you now.”
The technology is implemented, but nobody explains to the team why. Employees fear for their jobs, boycott the tool, or use it half-heartedly.
What happens: The AI solution is technically deployed but humanly rejected. Adoption rate: under 20%. ROI never materializes.
According to a Gartner study (2024), 87% of all digital transformation projects fail due to lack of user acceptance, not technical problems. The rate for AI projects is similar.
How to do better:
- Communicate early: AI doesn’t replace jobs, it replaces tedious tasks. Show concretely what improves for each individual.
- Train, train, train: Not once, but continuously. People need repetition and practice.
- Identify champions: Find 2-3 enthusiastic employees who serve as internal ambassadors.
- Take feedback seriously: When the team says “this doesn’t work like that,” they’re usually right.
A craft business in Bern with 25 employees got this right: before the new AI-assisted quoting system was introduced, three sales team members had shaped the tool over 6 weeks. Adoption rate at launch: over 90%. The result: quote creation time reduced from 45 minutes to 12 minutes.
The Approach That Works
I call it the Pragmatic AI Cycle:
1. IDENTIFY PROBLEM
A concrete, measurable business problem
|
v
2. FIND QUICK WIN
The simplest AI solution that makes a difference
|
v
3. PILOT (4-8 weeks)
Small team, clear scope, fixed deadline
|
v
4. MEASURE
Hard numbers: before vs. after
|
v
5. SCALE OR STOP
Honestly evaluate, then decide
|
v
6. NEXT PROBLEM
Back to step 1 -- with more competence and trustNo 100-page strategy document. No million-dollar budget. No external consultant who analyzes for 6 months and then leaves.
Instead: Start pragmatically, learn fast, improve measurably.
Comparison: Big-Bang vs. Pragmatic Cycle
| Criterion | Big-Bang Approach | Pragmatic Cycle |
|---|---|---|
| Timeline | 12-18 months | 3 months per cycle |
| Initial investment | CHF 50,000+ | CHF 500-5,000 |
| Measurable results | After 12+ months | After 8 weeks |
| Team burden | Very high | Moderate |
| Risk of failure | Very high | Low |
| Learning curve | Steep, at the end | Continuous |
Conclusion
AI in SMEs doesn’t fail because of technology. It fails because of:
- Missing problem focus (technology before need)
- Too-large scope (everything at once instead of one at a time)
- Neglected people (deploying tools without guidance)
The good news: All three traps are avoidable. You don’t need a data scientist or a million-dollar budget. Just a clear head, a concrete problem, and the courage to start small.
The SMEs that have understood this are already gaining measurable competitive advantages today, without taking large risk bets.
Want to introduce AI in your SME pragmatically, without falling into these traps? Let’s discuss.