Master Your AI Transformation – From AI Silos To T...

This episode is a follow-up to an earlier conversation with Patrick Kohler on AI strategy and scaling. Since that recording, the landscape has shifted noticeably: Agentic AI has moved from a future concept to an active discussion in boardrooms and transformation programs. This episode picks up where the last one left off – with more concrete frameworks, updated observations from client projects, and a sharper focus on what separates organizations that scale from those that stay stuck in pilots. (For a german speaking version listen in on Apple Podcasts · Spotify – or read the summary below).

Patrick Kohler, Head of AI Strategy and Advisory at SAP is focusing on how organizations can move from isolated AI projects to a scalable, strategically anchored Agentic AI architecture.

Why Agentic AI Is Not Just Another Hype

The fundamental difference between previous AI solutions and Agentic AI lies in the capacity to act. Classic Generative AI systems respond to requests – they deliver summaries, write emails, generate reports or find statistical patterns. Agents, by contrast, act actively: they can take non-deterministic steps, access tools and data, interact with other systems and people, and independently handle exceptions and deviations.

“Agents are proactive systems – they can overcome challenges autonomously, access tools and data, and interact across silos. That is exactly where most inefficiencies hide.”

That is precisely where the real potential lies: in most enterprise processes, it is not the standard cases that consume time – it is the exceptions, the cross-departmental coordination, the situational adjustments. Agents can address these points. And in doing so, they transform an ERP system from a passive data-entry tool into an active nervous system that executes processes while keeping the human in the role of orchestrator.

Key Numbers from the discussion

📊Numbers and Data

Metric Figure

AI project failures rooted in people, process & org 70%

AI project failures rooted in technology 30%

Process improvement via Retrofit 5–10%

Process redesign potential via Re-Engineering 60–90%

RFQ process duration – before AI Usage 20 days

RFQ process duration – after AI Usage 1 day

Invoice intake automated via Document AI 70%

These reflect patterns Patrick and his team observe consistently across client engagements.

Why Nine Out of Ten AI Projects Fail – And What to Do About It

Only five percent of companies achieve AI value at scale. The reasons are well-documented: missing goal definitions and KPIs, a too strong technology focus without a clear business value, the sheer pace of innovation that makes it difficult to distinguish signal from noise – and insufficient enablement of the people involved.

Many organizations start with a use case that deploys the latest LLMs and platforms but addresses no concrete business problem. The result: pilot purgatory. Projects that never reach scale. Patrick's observation from the field aligns with what applied research consistently shows – the organizations that break through share a common set of behaviors, which he calls a “strategic AI playbook.”

Agent Days-Intro.jpg

An overview on elements of an AI Playbook recently presented by Patrick on SAP Agentdays in Berlin.

What Leading Companies Do Differently

First, they define a clear AI strategy from the outset, directly linked to the overall business strategy. AI strategy and business strategy are, in Patrick's view, no longer two separate topics.
Second, they do not manage isolated use cases, but run a dedicated AI program with top-down commitment from C-level, a sound governance structure, and a continuous roadmap.
Third, they rethink processes – not just automate them. This distinction – retrofit versus re-engineering – is one of the most consequential decisions an organization can make, and it is covered in detail below.
Fourth, they invest in their people. Enablement is not a nice-to-have but a prerequisite: the better the people are trained, the better the use cases become.

Retrofit vs. Re-Engineering: A Decision That Defines Your Return

This is one of the most underestimated distinctions in AI transformation. The choice between the two approaches determines not just the efficiency gain, but whether the investment delivers genuine transformation or merely incremental improvement.

Dimension	Retrofit	Re-Engineering
Approach	New technology on existing process	Process redesigned around agents
Effort	Lower upfront	Higher upfront
Time to first value	Faster	Slower initially
Efficiency gain	5–10%	60–90%
Scalability	Limited	High
Risk	Lower short-term	Requires change management
Best suited for	Stable, low-complexity processes	High-frequency, exception-heavy processes
Human role	Largely unchanged	Shifts to orchestrator

Patrick's analogy captures the retrofit risk well: placing a Ferrari engine into a golf cart – you accelerate, but not safely, and the first accident is not far off. Re-engineering means asking, already at the process design stage: which steps should an agent handle? Where do I need a human in the loop? Where is my risk threshold? These questions are far more costly to answer retrospectively.

The Human-AI Interface: From Executor to Orchestrator

Patrick is clear on this point: humans are not being replaced – they are taking on a different, strategically more significant role. Moving away from data entry and passive execution, toward the orchestration of agents – comparable to a captain steering a ship while multiple systems operate in parallel.

“The human becomes the agent orchestrator – moving away from pure execution and data handling, toward truly strategic roles.”

A thought Patrick introduces: every employee becomes, in a sense, a people manager – with agents as new team members that need to be onboarded, monitored, and replaced when their lifecycle is complete. This requires knowledge of what AI can and cannot do, clear intervention thresholds, and – critically – that these questions are built into process design from the start, not added as an afterthought.

The concept of the “Circle of Control” serves as a useful guiding principle here: the more autonomy an agent receives, the more governance is required.

Common Misconceptions Worth Addressing

These are patterns Patrick and his team encounter regularly in client collaborations. Each one is worth examining directly.

“AI is primarily a technology problem.”
The data says otherwise. Seventy percent of AI project failures originate in people, processes, and organization. Technology accounts for only thirty percent. Most organizations, however, invest their attention in exactly the wrong proportion.

“A successful PoC means we are ready to scale.”
The step from pilot to productive deployment is one of the most frequently underestimated transitions in AI programs. Governance, change management, and scalability need to be planned from the very beginning – not figured out after the pilot succeeds.

“Agents will replace my team.”
Agents handle routine, repetitive, and exception-prone tasks. They free people to focus on judgment, domain expertise, and strategic decisions. The role changes; the human does not disappear.

“More AI tools equals more value.”
Tool proliferation without strategic alignment produces exactly the silos the transformation is supposed to eliminate. Value comes from integration, not accumulation.

“Every process needs an agent.”
Patrick is explicit: not every process benefits from an agent. Sometimes a simpler automation tool, an ML model, a digital workflow or even an unchanged process is the right answer. The starting point is always the business problem, not the technology.

Flywheel Effects: When Scaling Actually Happens

A flywheel effect does not emerge automatically – it requires three preconditions: a continuous AI program with clear governance as the foundation; modular, reusable components – Patrick cites Document AI as an example deployed both in procurement (RFQ process reduced from 20 days to 1 day) and in accounts payable (70 percent of invoice intake automated); and the human as a multiplier.

When one business area experiences tangible value, demand and knowledge exchange spread organically to other areas. Communities, internal networks, and shared learning are not soft add-ons – they are structural components of a functioning flywheel.

The SAP Approach: AI Center of Excellence and Strategic Program

Patrick describes a three-step approach his team applies with customers: first, a baseline assessment using AI Discovery Workshops and Maturity Assessments. Then a focus on business value through process analysis and prioritized use cases. Finally, the translation into a structured roadmap – not a wish list, but a plan with clear technological and data dependencies.

“Think of AI as a team sport – build a strong partnership and jointly identify where Business AI makes sense and where it creates real value.”

The AI Center of Excellence bundles these elements and helps organizations make the critical step from pilot to productive use. It is not simply a team of AI specialists – it is a structured organizational capability that connects business knowledge, process expertise, and AI technology in a continuous and consistent way.

Unpopular Opinion on AI Strategy

Patrick speaks of “agent washing” – analogous to the earlier “Gen AI washing”: the tendency to label everything as an agent simply because the term currently dominates. His position: Agentic AI is a valuable tool, but not always the right one. Those who start from the business problem choose the tool that solves it most efficiently – and that does not necessarily have to be an agent.

Inspiring Statements

“70 percent of the reasons why AI projects fail lie in people, processes, and organization – not in technology.”

“AI must be understood as a continuous journey, because the pace of innovation is significantly higher than with previous technologies.”

“Investing millions of euros and placing the latest technology on top of an old process – you shouldn't be surprised if you don't unlock the full transformation potential of AI.”

“The human becomes the agent orchestrator – moving away from pure execution and data handling, toward truly strategic roles.”

“Think of AI as a team sport – build a strong partnership and jointly identify where Business AI makes sense and where it creates real value.”

Practical Tips around AI Strategy

1. Align AI with Business Strategy
Do not treat AI as a separate IT topic. Link your AI ambition directly to your business objectives and define concrete KPIs that allow you to measure success – and to recognize when a use case is not delivering.

2. Set Up a Dedicated AI Program – Not a Project, Not a Silo
Instead of individual use cases, build a structured program with governance, C-level sponsorship, and a continuous pipeline. Only then does the flywheel effect emerge.

3. Start From the Business Problem, Not the Technology
Before deciding whether an agent, an ML model, or a classic automation tool is the right fit: define the pain point in the process first, then choose the appropriate tool.

4. Make a Conscious Choice Between Retrofit and Re-Engineering
For each process, assess whether optimization is sufficient or whether a fundamental redesign holds greater potential. Use ideation workshops with clear criteria: What data is available? Which KPIs should improve? Where is human in the loop necessary?

5. Plan the Step From Pilot to Production From the Start
The transition from POC to productive use frequently fails because it is planned too late. Build governance, scalability, and change management into the use case definition from the very beginning.

6. Try It Yourself – Don't Just Theorize
Patrick uses N8N and SAP Joule Studio to build agents himself. Hands-on experience helps develop realistic expectations and leads to better decision-making.

7. Use Notebook LM as a Learning Tool
Have studies and reports turned into podcasts and interact with the content directly – an efficient format for staying broadly informed in a fast-moving field.

Summary

Anyone serious about AI transformation cannot avoid three things: a clear strategy with measurable goals, a structured program instead of isolated pilots, and the consistent involvement of people – not as a risk factor, but as the central shapers of the process. Agentic AI is not hype, but a tool with real potential – as long as it is approached from the business problem outward. Check out also the shownotes for further information and contact us if you want to reflect and craft your business AI strategy.

Glossary from the podcast

Agentic AI – AI systems that act proactively, execute multi-step tasks autonomously, access tools and data, and handle exceptions without deterministic programming.

Agent Washing – The tendency to label any AI feature as an “agent” for marketing or relevance purposes, regardless of whether it actually exhibits autonomous, goal-directed behavior.

AI Center of Excellence (AI CoE) – A structured organizational capability that connects business knowledge, process expertise, and AI technology to govern, scale, and continuously improve AI adoption across an enterprise.

Circle of Control – A governance principle stating that as agent autonomy increases, the level of human oversight and governance required must increase proportionally.

Flywheel Effect – A self-reinforcing cycle in which early AI successes generate organizational momentum, knowledge sharing, and demand that accelerate further adoption and value creation. The more data, the more

Human in the Loop – A design principle where human judgment, approval, or oversight is built into an AI-driven process at defined intervention points, particularly for high-risk or high-stakes decisions.

Pilot Purgatory – The state in which AI use cases remain permanently in the proof-of-concept phase, never reaching productive deployment or generating measurable business value at scale.

Re-Engineering – The fundamental redesign of a business process with agents in mind from the start, enabling efficiency gains of 60–90% by rethinking roles, steps, and human-agent collaboration.

Retrofit – The application of new AI technology to an existing, largely unchanged process, typically yielding incremental improvements of 5–10%.

Strategic AI Playbook – A set of organizational behaviors consistently observed in companies that successfully scale AI: clear strategy, dedicated program governance, process redesign, and systematic people enablement.