01 · Pitfalls
Why AI projects fail
The reasons for failure repeat themselves with remarkable regularity. Seven pitfalls cover the vast majority of cases.
1. No clear goal
“We want to try some AI” instead of “We want to shorten the offer process from 5 days to 24 hours.” Without a measurable goal, no measurable success.
2. Too many pilots
Eight pilots in parallel, each in a different area. Resources get stretched, learning does not occur, no one can say what worked.
3. IT instead of organization
IT introduces, the business does not accept. Without change management, AI remains a foreign body.
4. Governance forgotten
Who may decide what with AI? What happens after an AI error? Without guardrails, wild growth and compliance risk.
5. No impact measurement
Many teams don’t know after the pilot whether it succeeded – because no baseline was measured before.
6. Poor data foundation
AI works with the data quality it is given. Poor data produces poor output – and erodes trust.
7. Leadership not on board
If leaders don’t understand, challenge and use AI output, AI systems remain ineffective in daily work.
02 · Adoption model
Four phases to anchored impact
Each phase has a clear goal, defined outputs and quality criteria for the transition to the next.
- 01
Diagnostics & prioritisation (wk 1–4)
AI readiness assessment, inventory of existing initiatives, use-case identification and prioritisation. Output: prioritised use-case list with assessment matrix.
- 02
Piloting (wk 5–12)
Maximum three use cases in parallel. Measure baseline, set up AI system, train teams, integrate. Output: measured impact vs. baseline, documented learnings.
- 03
Anchoring (wk 13–20)
Move successful pilots into regular operations. Train leaders in AI steering. Operationalise governance. AI coordinator takes over.
- 04
Scaling (from month 6)
Whatever works is rolled out – adapted, not copied. The Early Warning Canvas – a structured monitoring tool combining AI impact, usage data and regulatory developments in one dashboard – ensures ongoing steering. → Discover the Canvas
| Phase | Critical success factors |
|---|---|
| Diagnostics & prioritisation | Executive commitment, cross-functional team, clear prioritisation rule |
| Piloting | Dedicated time budget, baseline measurement before start, governance frame defined upfront |
| Anchoring | Process documentation, leadership training, communication to all affected, change support |
| Scaling | Documented success model, resource plan, AI coordinator with clear remit, monitoring dashboard |
03 · Prioritisation
Use-case prioritisation in SMEs
The most important decision is not technology selection. It is use-case prioritisation: which two or three applications have the largest impact potential – and are implementable in the next three months?
| Criterion | Guiding question |
|---|---|
| Impact potential | How strongly does this use case change a key metric – time, cost, quality, customer satisfaction? |
| Data availability | Are the necessary data available, structured and of sufficient quality? |
| Process clarity | Is the process targeted for AI integration well documented and stable? |
| Compliance requirements | Is it a high-risk use case under the EU AI Act – and manageable? |
| Team readiness | Is the affected team ready and motivated to run an AI pilot? |
| Scalability | Can a success in this area be transferred to other areas? |
High impact
Offer creation
AI-supported drafting of offers and calculations. Time savings: 40–70 %. Requires structured product data.
High impact
Document analysis
Contracts, suppliers or compliance docs analysed automatically. Time savings: 50–80 % on first review.
High impact
Knowledge management
Internal AI assistant with access to manuals, processes, FAQs. Faster onboarding, scaling expert knowledge.
High impact
Reporting
Automated aggregation of reports and dashboards. Time savings for leaders: 3–5 hours per week.
High impact
Customer communication
First responses to standard inquiries, email classification, ticket routing. Without human approval only for clear low-risk cases.
High impact
Quality control
AI-supported review of outputs – in production, documentation or code. Reduce error rate, increase learning speed.
04 · Case studies
Two examples from practice
Two anonymised examples show what structured AI adoption can achieve in 10–14 weeks.
Engineering – fewer active pilots
Special machinery builder (220 FTE): focus from 12 down to 4 use cases. 100 % EU AI Act compliance. 4.5× higher adoption in target processes after 14 weeks. Detailed governance view on this case in the whitepaper „AI Governance for SMEs“.
Professional services – faster offers
Management consultancy (85 FTE): offer assistant, internal knowledge system, research automation. 10 weeks to an anchored AI strategy. −60 % research time. Full case study and implementation details: resilience-bakery.de/whitepaper
Key observation
Both cases started with the same problem: too many initiatives, too little impact. Both gained impact by reduction, not by adding more.
05 · Series & authors
About this whitepaper series
This document is part 5 of the Resilience Bakery whitepaper series on organisational resilience and AI in mid-sized companies.
Part 1
Organisational resilience
Six dimensions and the resilience trilemma as foundation.
Part 2
AI governance for SMEs
Roles, approvals, EU AI Act and high-risk use cases.
Part 3
Resilience maturity model
EFQM and TEMP framework, five levels from reactive to anticipatory.
Part 4
AI readiness
Six readiness dimensions and the path to impact.
Part 5
AI adoption for SMEs
You are reading this part – four phases, seven pitfalls, two case studies.
Authors
Malte Lohrberg & Ralf Metternich
Resilience Bakery supports mid-sized companies in organisational resilience and structured AI adoption – through diagnostics, advisory and training. Our services: Resilience Check · Path to Resilience · AI as Enabler · Transformation & Enablement · Training. Locations: Düsseldorf & Frankfurt · resilience-bakery.de
All whitepapers
Overview of all parts and PDF downloads: resilience-bakery.de/en/whitepapers