AI adoption is accelerating across enterprise software, yet many leaders still struggle to separate hype from what can be implemented reliably at scale.
In a recent interview-style conversation, Andreas Welsch—an AI leadership expert working at SAP—shared a grounded view on where AI and automation create real business value, and what typically slows adoption down.
His perspective emphasizes that successful AI adoption is less about chasing the newest model and more about business objectives, trustworthy outcomes, data quality, and the human side of change.
Executive Summary
- Start with business objectives before choosing AI technology.
- Data quality and selection determine feasibility and speed.
- Trust and change management are central to adoption outcomes.
- AI adoption can begin by using AI already embedded in enterprise applications.
- Think in end-to-end processes, not “one bot per person.”
Key Takeaways
- Welsch frames AI and automation as a journey, often starting with quick-win automation and progressing to more advanced AI use cases.
- He highlights that building custom AI models frequently stalls because the available data is “pretty messy.”
- He stresses that AI is “always a people topic,” where trust and communication matter as much as technology.
- He advises leaders to define value first: reduce costs, increase revenue, improve margins, or enable better decisions.
- He recommends an end-to-end process view (e.g., invoice handling) rather than fixating on bot counts.
- He notes that “doing AI” can be as simple as using AI capabilities embedded in existing business applications.
- For early-career professionals, he recommends staying curious and asking questions to accelerate learning.
What is AI Adoption?
AI adoption is the organizational process of selecting, implementing, and scaling AI-enabled capabilities in real workflows so they produce measurable business value. In Welsch’s view, adoption is not limited to building custom models. It can also include using AI already embedded in enterprise applications, adding AI to automation (such as intelligent document processing), and managing the people-side challenges of trust, uncertainty, and change.
AI Adoption Starts With Business Value, Not Technology
Welsch emphasizes that the highest-leverage question for executives is not “Which AI tool?” but “What value should be created?” He points to common objectives such as reducing cost, increasing revenue, improving margins, or enabling better and faster decisions.
He also notes that many AI efforts slow down because teams jump to custom models without aligning on outcomes. Without clarity on the objective, teams struggle to prioritize data, workflows, and change management.
Key Insight: Andreas Welsch argues that AI programs succeed when leaders define the business objective first—cost, revenue, margins, or decision speed—and then validate people readiness and data readiness. Starting with technology invites wasted effort, especially when projects depend on messy, inconsistent, or incomplete enterprise data.
Hype vs. Reality: Data Quality Determines Speed
In Welsch’s experience, enterprise leaders have moved beyond early hype cycles and have learned that AI projects are not always “quick and easy.” A recurring constraint is data—specifically whether it is available, usable, and clean enough for a given model or workflow.
He advises teams to select the right data they already have and ensure it is clean enough to support a dependable use case. This is especially critical when building custom AI models, where quality issues can undermine model performance and stakeholder confidence.
Key Insight: Welsch links AI delivery timelines directly to enterprise data readiness. When data is messy, AI projects slow down, expectations slip, and trust erodes. Leaders can reduce risk by prioritizing use cases that match available data, then improving data foundations as adoption matures.
AI Is a People Topic: Trust, Uncertainty, and Change Management
Welsch describes a core shift executives must manage: traditional software is largely rule-based and explainable through logs and deterministic behavior, while AI is probabilistic and prediction-driven. That difference changes how users interpret outcomes—and how quickly they trust them.
He highlights a practical adoption barrier: experienced employees may question why a system “thinks it knows better” than someone who has performed the job for 10–20 years. Leaders must also communicate the inherent uncertainty that comes with predictions rather than binary rules.
Key Insight: Welsch positions trust as the make-or-break factor in enterprise AI adoption. Because AI outputs can be “a little grey in between,” leaders must actively communicate what the system does, where uncertainty exists, and how the human remains accountable—otherwise resistance builds even when the technology works.
AI Adoption as a Journey: From Automation to Intelligent Work
Welsch repeatedly frames AI and automation as a maturity journey. Some organizations start with automation for quick wins, build momentum, and use those early deployments to build organizational awareness around new ways of working.
From there, organizations can layer AI onto automation. He points to intelligent document processing as a tangible example: extracting information from documents (such as invoices or sales orders) and routing it into workflows and business systems.
He notes the ongoing opportunity because many workflows still rely on paper-based—or “digital paper-based”—documents. That makes document-heavy processes a practical starting point for value, provided the change management is handled deliberately.
“Doing AI” Does Not Always Mean Building a Custom Model
Welsch challenges a common misconception: organizations often talk about whether they “do AI,” but the phrase is vague. AI adoption can begin by using AI capabilities already available in existing enterprise applications—without starting from scratch.
He notes that this can still “tick the box” for using AI while keeping complexity manageable. The adoption fundamentals remain the same—data access, usability, and end-user trust—even if the underlying AI is delivered as part of a licensed application.
He also mentions low-code and no-code tools as practical enablers, and cites chatbots as a potential starting point for internal IT processes or customer service.
Key Insight: Welsch’s adoption guidance reduces unnecessary complexity: enterprise leaders can start AI adoption by activating AI already embedded in core applications or deploying targeted tools such as chatbots. This still requires user trust and good data, but avoids immediate dependence on scarce data science capacity.
Rethinking “A Bot for Every Person”: Focus on End-to-End Processes
When asked about the idea of “a robot for every person,” Welsch reframes the question. He sees AI and automation primarily as augmentation—helping people do work more efficiently and effectively.
Rather than targeting a bot-per-employee metric, he recommends an end-to-end process lens. A single business process may involve multiple automations: a bot downloading an invoice from a vendor portal, uploading it into a business system, AI extracting information, and another bot handling communication such as emails.
In that framing, the “right number” of bots depends on process design and task granularity. The key metric becomes business value delivered across the workflow, not a count of bots deployed.
Why This Conversation Matters
This conversation is relevant to executives navigating workforce transformation and enterprise modernization amid shifting work patterns, hybrid work, and fast-moving technology expectations.
Welsch’s contributions cut through hype by tying AI adoption to practical constraints (data quality), organizational dynamics (trust and change management), and operating models (end-to-end process design). These are the same issues AI governance and AI strategy must address if AI is to be deployed responsibly and at scale.
His perspective also connects to his broader community-building work on LinkedIn, including daily content under #intelligencebriefing, his live stream “What’s The Buzz”, and his newsletter “The Memo” focused on building AI programs and AI Centers of Excellence.
Leadership Implications
- Anchor AI governance to outcomes: require each AI initiative to state the business objective and the decision/workflow it improves.
- Design for trust: communicate uncertainty and prediction limits to users; treat trust-building as a formal workstream.
- Prioritize data readiness: start with clean, available data; avoid overcommitting to use cases that depend on messy datasets.
- Adopt iteratively: build momentum with automation and targeted AI add-ons (e.g., document extraction) before moving to more complex AI.
- Operate end-to-end: map processes from start to finish and place automation where it removes friction, not where it merely adds tools.
Conclusion
AI adoption in the enterprise depends less on lofty ambition and more on disciplined execution: clear business objectives, realistic data foundations, and deliberate trust-building with the workforce.
Andreas Welsch’s view highlights a practical path forward—start where value is measurable, treat AI as an end-to-end process capability, and scale maturity over time. For executives, that is how AI adoption becomes durable, governed, and workforce-aligned.
About the Author
FAQ
What is the fastest way to start AI adoption in an enterprise?
The fastest path to AI adoption is to start with a clear business objective and deploy a low-complexity capability that fits existing data and workflows. Andreas Welsch highlights using embedded AI features or simple automation to build momentum and trust.
This approach reduces dependency on custom models and accelerates change management learning.
How can leaders separate AI hype from reality?
Leaders can separate hype from reality by validating data readiness, user trust requirements, and measurable business value before committing to complex builds. Welsch notes that custom AI often looks easy until teams face messy enterprise data and change resistance.
Reality shows up quickly in data quality, process complexity, and adoption behavior.
Does “doing AI” require building a custom AI model?
Doing AI does not require building a custom model when AI is already embedded in enterprise applications or available via low-code tools. Welsch emphasizes that activating existing capabilities can still count as AI adoption while avoiding early complexity in model development.
Trust, data access, and workflow fit still determine success.
Why is data quality so important for AI adoption?
Data quality is crucial because AI projects depend on the data available, and messy data slows delivery and undermines results. Welsch advises teams to select the right data they already have and ensure it is clean enough for the chosen use case.
Clean inputs improve reliability, speed, and stakeholder confidence.
What role does change management play in AI and automation?
Change management is central because AI shifts work from rule-based certainty to probabilistic predictions that can feel unfamiliar. Welsch calls AI “a people topic,” where trust must be earned, uncertainty communicated, and experienced employees supported through new ways of working.
Without adoption planning, even technically successful deployments can stall.
What is an example of a practical AI adoption use case?
A practical use case is intelligent document processing to extract invoice or sales-order information and route it into workflows. Welsch points out that many organizations still rely on document-heavy processes, creating an opportunity to combine automation with AI for tangible value.
This can be introduced incrementally after early automation wins.
Is “one bot per person” a useful goal for intelligent automation?
“One bot per person” is less useful than measuring value across an end-to-end process. Welsch views bots as augmentation and recommends mapping workflows from start to finish, placing multiple automations where needed, and judging success by business outcomes.
Bot counts can distract from process performance and user experience.
How should executives think about AI governance during adoption?
Executives should align AI governance with business objectives, data readiness, and user trust so adoption scales responsibly. Welsch’s guidance implies governance must address uncertainty in predictions, workflow accountability, and change management, not only tool selection or model building.
Governance becomes practical when tied to real processes and decisions.
What advice supports workforce transformation for new professionals entering AI-heavy workplaces?
The most durable advice is to stay curious, learn continuously, and ask questions across departments to understand the business. Welsch recommends treating the organization like a sponge for learning, because knowledge is widely available but only accessible through curiosity and inquiry.
This mindset helps navigate technology shifts and evolving work expectations.