AI leadership is increasingly less about chasing the next model release and more about aligning people, data, and decisions around real business outcomes. In a wide-ranging conversation recorded at SAP’s Experience Center in Newtown Square, AI leadership expert Andreas Welsch explains why most AI programs stall: organizations treat AI as an add-on instead of a catalyst to rethink work. Andreas Welsch—an SAP alumnus with 23 years in the ecosystem, including eight years in machine learning and AI—now advises leaders across industries on AI strategy, roadmaps, and adoption. His focus is pragmatic: cut through vendor noise, ground efforts in business strategy, and bring the workforce along.
Source context: The insights below are normalized from a podcast-style executive discussion on Transformation Every Day with host Alexander Greb, featuring Andreas Welsch.
Why this conversation matters
This discussion reflects what many executives face right now: weekly AI headlines, rising expectations from boards and vendors, and early signs of “AI fatigue.” Welsch frames the moment as a leadership and workforce transformation challenge, not a technology procurement exercise.
The audience includes CIOs, CTOs, CHROs, enterprise architects, and transformation leaders—especially those navigating SAP modernization—who need a clear path from AI possibilities to operational reality.
Executive Summary
- AI leadership starts with business strategy, not “top AI use cases.”
- Agentic AI is promising, but adoption will lag due to trust, data, and controls.
- Culture determines whether employees use AI—or hide usage from managers.
- Data readiness still gates success, even with generative AI and RAG approaches.
- Roles will change; adaptability becomes a core career skill.
Key Takeaways
- Welsch advises leaders to begin with where the business wants to go, then map AI to outcomes.
- He expects AI vendors and enterprise adoption to stay out of sync, creating hype cycles and fatigue.
- He warns that many organizations are “very badly prepared” on data visibility and access controls.
- He sees agentic AI as a step-change for research, knowledge retrieval, and delegated work.
- He argues cloud environments provide practical access to AI capabilities, scale, and model choice.
- He predicts more automation in areas like testing and optimization—shifting human work to review and guidance.
- He highlights adaptability and “skating to where the puck is going” as future-proof skills.
What is AI leadership?
AI leadership is the discipline of turning AI capability into business outcomes by aligning strategy, governance, culture, and workforce enablement. In Welsch’s view, technology is necessary but insufficient: leaders must secure buy-in, address data foundations, manage risk (privacy and security), and create an environment where employees feel safe to adopt AI in daily work. Effective AI leadership also means filtering vendor noise, choosing what is usable today, and preparing for emerging shifts such as agentic AI—software that can act with delegated autonomy across tasks.
From SAP AI to independent AI strategy advisor
Welsch spent 23 years at SAP, including leadership work spanning Leonardo, AI in S/4HANA, and SAP’s marketing organization for Business AI. He left SAP after seeing a broader market gap: leaders across industries feel pressured to “do something with AI,” but lack clarity on where to start and how to bring people along.
As an independent advisor, he now views AI through an operator lens: responsible for marketing, revenue, and finance in his own business. That perspective sharpens his focus on systems that automate repeatable work, reduce cycle time, and scale output without sacrificing quality.
Key Insight: Welsch’s shift from vendor to advisor highlights a recurring AI leadership gap: leaders rarely need “more AI features”—they need a roadmap that ties AI to strategy, adoption, and operating reality, including training, security, and data readiness.
AI in practice: automation that scales a single executive
Welsch describes a concrete workflow automation: converting a livestream into a podcast episode, newsletter article, and social posts. Using make.com for orchestration, OpenAI’s GPT-4o model for content generation, and Stability AI for an image, he reduced a two-to-three-hour process to roughly a minute and twenty seconds.
He distinguishes between thought leadership and “AI workslop.” The source ideas should remain human-generated and differentiated; AI’s role is repackaging and distribution across formats, with human review to ensure accuracy and voice alignment.
Key Insight: This is a practical model for AI adoption: keep human judgment and originality where it matters, then use AI to industrialize repetition. For executives, the value is not novelty—it is throughput, consistency, and reclaimed time for higher-value decisions.
Agentic AI: why the hype is real—and why adoption will lag
Welsch calls agentic AI the “new hype topic,” but he also points to tangible near-term value: agents that research the week’s most important topics, synthesize key articles, or retrieve answers from a repository of prior content. For creators and executives, this reduces manual research and accelerates preparation.
He also cites emerging distribution models: voice cloning and AI avatars. Using 11 Labs, he created an AI voice clone by recording roughly 45 minutes of sample text, then produced an audiobook for about $100—contrasting with $5,000–$6,000 and longer timelines for professional voice acting. He also references avatar platforms such as HeyGen, Synthesia, and D-ID, and expects licensing models where creators monetize voice and likeness rather than on-camera time.
Key Insight: Agentic AI is not only a productivity layer; it alters operating models. When “delegation to software” becomes normal—research, drafting, scheduling, and knowledge retrieval—leaders will need governance, controls, and new skills for supervising agent output.
Noise vs. signal: grounding AI leadership in what works today
Welsch notes that headlines now arrive weekly: claims about building AGI, near-term “agents in the workforce,” model releases, and controversies around guardrails. He advises enterprise leaders to recognize the pace—but stay grounded in what can be used today to solve real business problems.
He also frames two forces moving at different speeds: vendors shipping new AI capabilities and companies adopting them. Enterprises must manage people, processes, procedures, security, and privacy—constraints that make adoption inherently slower than model innovation.
Key Insight: AI leadership requires a “signal filter.” The strategic advantage comes from disciplined evaluation and adoption—pilots, controls, training, and measurable outcomes—not from reacting to every new model announcement.
AI washing and the foundation problem: data, access, and security
Welsch acknowledges “AI washing,” where vendors subsume rule-based logic, machine learning, and generative AI under one marketing label. For customers, the practical response is due diligence: look “under the hood,” clarify what is truly AI-driven, and evaluate fit for the organization’s needs.
His sharper warning is about readiness. When organizations deploy AI that can search or summarize, hidden information can surface—such as documents or pay-related information not intended for broad visibility. Welsch’s assessment is blunt: many organizations are “very badly prepared.”
Even with generative AI approaches such as retrieval augmented generation (RAG), data remains the gate: clean, fresh, complete data; clear access controls; and security and privacy guardrails. Without these foundations, AI outputs degrade—or become risky.
Culture is the hidden control plane of AI adoption
Welsch argues that AI adoption hinges on culture and communication, not software availability. He cites a Slack workforce analysis: 46% of participants said they did not tell managers they used AI at work, fearing they would appear lazy or incompetent—or simply receive more work.
That finding points to a leadership failure mode: deploying AI tools without creating psychological safety and shared expectations. Leaders must encourage experimentation—even if it takes longer at first—because employees need to build “muscle memory” for interacting with AI tools effectively.
Key Insight: When employees hide AI usage, governance becomes impossible and benefits stay fragmented. AI leadership should normalize approved tools, define safe usage, and reward learning—so adoption becomes visible, measurable, and improvable.
Where enterprises should start: strategy first, use cases second
Welsch pushes back on the common question, “Where should companies start?” In his view, asking for “the top AI use cases” is like getting a root canal when the real need is a cleaning—or when there is no problem at all. A generic answer misaligns effort and outcomes.
Instead, he recommends beginning with business strategy: revenue growth plans, new product lines, and targeted cost reduction. AI should enable those goals. From there, leaders can evaluate practical embedded-AI capabilities already available across software, including meeting transcription, document processing, contract comparison, and drafting.
He also expects a familiar adoption pattern: early adopters show proof points, then peers feel pressure as competitors cite cost reductions or performance improvements. AI fatigue can quickly flip into urgency—often too late for careful strategy.
Cloud vs. on-prem: why deployment choices shape AI velocity
Welsch states that AI can work for many organizations, but foundations and environment matter. Cloud enables scale and choice—such as consuming model APIs rather than buying costly infrastructure (including specialized hardware). For many midsize companies, that difference is what makes an AI business case feasible.
He contrasts cloud-enabled consumption with the realities of on-prem investment, while acknowledging that different environments serve different purposes. The executive question becomes speed and innovation: what enables faster evaluation, safer adoption, and practical ROI?
Workforce transformation: what changes for consultants and architects
Welsch expects roles to change as AI becomes reliable, enterprise-grade, and cheaper than manual effort. For example, he references early signs of AI-driven optimization in SAP landscapes and anticipates a shift where humans may no longer perform some manual testing tasks—but instead instruct tools, review outputs, and ensure quality.
He frames this as inevitable change: tasks evolve, and many professionals will become “leaders of a team,” potentially supervising agents rather than only humans. The priority skill becomes adaptability—paired with anticipating change, not merely reacting to it.
He recommends starting with hands-on usage of AI tools, even personally, to understand capabilities and shortcomings. That experience translates into stronger advisory capacity when helping customers redesign workflows and processes.
Leadership Implications
- Anchor AI leadership in business strategy: define growth and efficiency goals, then map AI initiatives to outcomes.
- Invest in data foundations: prioritize data quality, freshness, completeness, and access controls before scaling AI.
- Make adoption safe and visible: set cultural norms so employees do not hide AI usage; standardize approved tools.
- Design workflows, not demos: incorporate AI into end-to-end work, with human review points and accountability.
- Prepare for agent supervision: develop skills and governance for instructing, validating, and auditing agentic AI outputs.
How this connects to Andreas Welsch’s broader work
Welsch’s perspective is consistent across the conversation: successful AI adoption depends on leadership beyond technology. He points to governance, stakeholder alignment, storytelling, culture, collaboration, and use case prioritization as the often-missed requirements for durable results.
He also references his book, The AI Leadership Handbook, which synthesizes insights from more than 60 AI leaders he interviewed on his live stream and podcast What’s the Buzz. He also notes his weekly LinkedIn Live session and his YouTube channel, Intelligence Briefing, as ongoing executive education channels.
Conclusion
AI leadership is the differentiator between experimenting with tools and transforming how work gets done. In Welsch’s view, agentic AI, voice cloning, and embedded AI features are accelerating, but outcomes still depend on strategy, data readiness, governance, and culture.
Executives who treat AI as a workforce transformation—built on foundations and adoption design—will move faster with less risk than those who chase hype cycles and bolt AI onto legacy practices.
FAQ
1) What should executives prioritize first in AI leadership?
Executives should prioritize business strategy alignment before selecting AI tools or use cases. Welsch emphasizes starting with where the company wants to go—growth, new products, or cost reduction—then identifying how AI enables those outcomes.
This avoids “random use case lists” that waste time and create fragmented adoption.
2) Why do many AI programs stall after early pilots?
Many AI programs stall because adoption is slower than vendor innovation and the enterprise foundations are missing. Welsch points to people, processes, security, privacy, and data quality as the constraints that prevent scaling beyond pilots.
Without governance and workforce enablement, AI remains a demo rather than an operating model change.
3) What is agentic AI and why is it relevant now?
Agentic AI refers to AI agents that can take delegated actions—such as researching topics, synthesizing information, or retrieving answers from a content repository—rather than only generating text. Welsch calls it a major new hype topic with practical near-term value.
Relevance grows as leaders seek automation that reduces research and coordination work.
4) How can leaders reduce “AI fatigue” in their organizations?
Leaders can reduce AI fatigue by filtering noise and focusing on what is usable today to solve real business problems. Welsch recommends staying grounded, avoiding constant reaction to headlines, and using AI only where value is clear and measurable.
This shifts attention from hype cycles to outcomes and adoption design.
5) What are the biggest risks of enterprise AI without proper preparation?
The biggest risks include exposing information that should not be widely visible and producing unreliable outputs due to weak data foundations. Welsch says many organizations are badly prepared on access controls, privacy, and data quality required for safe AI adoption.
These risks increase as AI gains search and summarization capabilities across content.
6) Does generative AI reduce the need for data quality work?
No—generative AI does not remove the need for clean, fresh, complete data. Welsch notes that approaches such as retrieval augmented generation still depend on organizational data, meaning data readiness remains a prerequisite for high-quality, tailored, and safe outputs.
Data work may be less glamorous, but it still gates AI outcomes.
7) Why do employees hide AI use at work?
Employees hide AI usage when culture punishes experimentation or signals that AI use implies laziness or incompetence. Welsch references a Slack analysis where 46% avoided telling managers they used AI, also fearing they would simply receive more work.
AI leadership must create psychological safety and explicit norms for approved usage.
8) Will AI replace consultants and enterprise architects?
AI is likely to change, not instantly eliminate, many consulting and architecture tasks—especially as outputs become reliable and cheaper than manual effort. Welsch expects shifts toward instructing tools, reviewing results, and advising on redesigned workflows rather than performing repetitive work.
Adaptability and the ability to supervise AI outputs become key differentiators.
9) What skills should leaders and professionals build for an AI-driven future?
The core skill is adaptability—paired with anticipating change. Welsch argues that people must build “muscle memory” by using AI tools directly, learning capabilities and limitations, and preparing to lead teams that may include AI agents performing delegated tasks.
This strengthens decision quality and keeps roles relevant as workflows evolve.