She wears two hats – and both fit snugly: professionally, Julia Schwäke leads Innovation & Go-to-Market for digital products at a German building materials manufacturer. Academically, she is doing her doctorate at HHL on the interplay of digital transformation, AI, and SME reality. In a conversation with Maximilian Hahnenkamp, she discusses data protection as a location issue, why many AI projects fail, how GenAI increases the cadence in startups – and why "just doing it" only works if the homework is done.

Research that reaches into the workshop

Schwäke's dissertation is paper-based: several studies on

• AI Implementation in SMEs (barriers, levers, maturity levels),

• Sustainability through AI (material use, energy, emissions),

• Managing Uncertainty (policy, funding, governance),

• Speed through GenAI – especially in startups.

"AI is no longer a nice-to-have. It is mandatory – across all company sizes."

Surprisingly for her: Political framework conditions play a much larger role for SMEs than often assumed – from data protection to funding logic.

Data Protection: A Brake or a Competitive Advantage?

Yes, Europe is stricter – and that temporarily slows down pace. But Schwäke argues pragmatically: Those who work responsibly with data build long-term trust, protection, and quality into their models. Her essence: Not "either-or", but "secure and fast – through clear rules".

Efficiency and Effectiveness: The "lost" Plus

GenAI accelerates code, texts, research – efficiency. The forgotten lever, says Schwäke, lies in effectiveness:

• For example, if developers can deliver faster thanks to AI, they can take over customer support, learn the pain points "on the wire" – and build more relevant products.

• AI therefore does more of the right thing, not just the same faster.

Why 95% Fail – and 5% Win

Schwäke's diagnosis is unambiguous:

1. No real use case

   "Golf course decision: 'We need AI. For what? No idea.'"
   Solution: Problem → Outcome → Data basis → KPI → Pilot → Scale.

2. Overselling & Broken Expectations
   Startups promise "learning AI", but deliver heuristics.
   Solution: Transparency about maturity level, scope, dependencies.

3. Missing Project Mechanics
   Without roles, data access, DQ checks, and milestones, governance fails.
   Solution: small steps (8–12 week pilots), clear "exit/expand" criteria.

4. Unfinished Foundations
   Process harmonization, data flow, metadata – homework first.

Corporate × Startup: How to Make it Work

• Expectation Management on both sides: What is deliverable today?

• Collaborative Escalation: "To achieve goal X, we need Y resources."

• Sell "Problem, not AI": Avoid buzzwords with users, cleanly explain benefits + governance to investors.

Where Immediate Value Lies (Julia’s Shortlist)

1. Production Optimization: Less material, less waste, more stable quality.

2. Bureaucracy & Back Office: Automated processes, less "monkey work".

3. Process Telemetry: Continuous data pathways for continuous improvement.

Where We Stand in the Hype Cycle

Schwäke locates us in the Valley of Disappointment: The euphoria gives way to reality, now the professional ascent begins – slower, but more robust. AI has come to stay.

7 Concrete Steps for the German Middle Class

1. Start with a Use Case Canvas: Problem, Benefits, Data, KPI, Risks.

2. Data Readiness: Source systems, data quality, access rights, logging.

3. Human-in-the-Loop mandatory: Quality, liability, learning in operations.

4. Piloting, not Postulating: 10-week pilot, strict go/no-go criteria.

5. Transparency in the Tech Stack: What is rule-based, what is ML/GenAI – and why.

6. Change by Design: Roles, training, communication plan, support.

7. Disclose ROI: Time savings and effectiveness gains (e.g., NPS, conversion, first-time-right).

Looking Ahead: Medicine, Personal Agents & Digital Hygiene

Her desired headline in ten years: "AI helps heal diseases – with humans in the loop."
For this, she names three future areas:

• Bioengineering & Personalized Therapies – AI as a diagnostic and design engine.

• Personal Agents – digital assistants that autonomously relieve everyday life and work.

• Energy-efficient AI & Fact Hygiene – green computing, better data standards, less fake.

"The simple problems are disappearing. Today's mediocrities will become easy. The hard problems of tomorrow we do not yet know."

Conclusion

Courage yes – but with method.
Those who today execute smart use cases cleanly, take Human-in-the-Loop seriously, and do their data homework, will reap not only efficiency but also strategic impact: better products, faster learning cycles, more resilient organizations.