Skip to content
Matthias Stegmüller
Validated with customers

Building a new approach to enterprise AI innovation.

AI agents open up enormous possibilities, but not every opportunity is worth pursuing. I lead SAP's cross-supply chain Agent Incubator, bringing together design, product, engineering, and enterprise customers to identify meaningful operational problems, explore agentic solutions, and rapidly prototype experiences that can be tested in real business contexts.

At a glance

  1. 01The mandate

    Start from the problem, not the technology

    Work directly with enterprise customers to understand complex business processes and translate opportunities into tangible, testable product concepts.

  2. 02My role

    Design leadership and product shaping

    I lead the design direction and discovery approach: which problems are worth solving, and how autonomous systems should take part in enterprise work.

  3. 03The method

    A repeatable way to find agentic value

    I developed a discovery and solution-design method that separates problems that need agentic AI from those better solved with conventional software.

01

Overview

What is the Agent Incubator?

The SAP Supply Chain Management Agent Incubator is a cross-functional innovation initiative focused on identifying, validating and prototyping high-value applications of agentic AI in enterprise supply chain operations. Its purpose is to bridge the gap between emerging AI capabilities and real operational problems.

Rather than starting with technology or predefined AI agents, the Incubator works directly with enterprise customers to understand complex business processes, identify opportunities for intelligent automation, and translate them into tangible, testable product concepts.

It brings together product design, AI engineering, product management, domain experts and customers to explore how AI agents can take responsibility for meaningful parts of enterprise work, while maintaining appropriate human oversight, transparency and control.

The ambition: move beyond isolated AI assistants and demonstrations towards integrated, outcome-driven AI systems that can operate reliably within real enterprise environments.

My work spans four connected areas

  1. 01

    Product strategy & opportunity discovery

    Identify high-impact supply chain problems, assess whether they suit agentic AI, and shape opportunities around measurable business outcomes rather than technological novelty.

  2. 02

    AI systems & experience design

    Define how people delegate work to AI, how agents coordinate and carry out tasks, where decisions need human intervention, and how execution stays observable, explainable and governed.

  3. 03

    Cross-functional leadership

    Lead and develop a multidisciplinary Incubator team, set design quality standards, and align customer needs, engineering feasibility, product direction and platform capabilities.

  4. 04

    Customer co-innovation & validation

    Facilitate discovery workshops with enterprise customers, turn operational complexity into executable scenarios, and guide rapid prototyping and validation in realistic system environments.

02

The method

From operational problems to agentic solutions

I developed a structured discovery and solution-design method so the Incubator can consistently move from ambiguous customer challenges to validated agentic opportunities. It answers one question: which problems genuinely benefit from agentic AI, and which are better solved with conventional software or automation?

The Agentic Solution Discovery Framework
  1. 01

    Frame

    Understand the operational problem

  2. 02

    Structure

    Map the business and decision process

  3. 03

    Qualify

    Assess agentic value and feasibility

  4. 04

    Prototype

    Design and implement an end-to-end execution slice

  5. 05

    Validate

    Test value, trust and operational fit

↺ Each stage produces evidence and decisions that inform the next. The process is iterative rather than strictly linear.

DiagramThe Agentic Solution Discovery Framework: five stages, each producing evidence and decisions for the next.

A shared blueprint for business, AI and human work. A central artifact of the method is an integrated execution blueprint that shows three dimensions of an agentic solution at once: the business process, what the agent executes, and where people are involved.

DiagramA delegation map for order fulfilment: one Job across five process stages, with the agents and workflow, systems, people and guardrails at each stage. Simplified example; the actual method also captures dependencies, decision boundaries, readiness criteria and validation activities.

The blueprint gives business stakeholders, designers, engineers and product teams a shared language. It keeps the solution from being reduced to either a process diagram or a collection of AI capabilities.

The method also sets explicit readiness criteria: is there a clear business outcome, an identifiable owner, feasible system access, meaningful automation potential and appropriate governance?

03

Customer engagement

Intelligent order fulfilment

A global manufacturer runs a complex make-to-order business, where reliable order confirmations depend on raw material availability, procurement lead times and existing production commitments.

The challenge. When urgent orders arrive or supply conditions change, planners have to weigh competing customer priorities by hand, find alternative sourcing, and sometimes reallocate material already committed to other orders. These decisions need coordination across planning, procurement and order management, which creates overhead and makes it hard to respond quickly without breaking existing commitments.

The opportunity. We explored how an agentic system could continuously evaluate material constraints, identify at-risk commitments and proactively recommend feasible resolutions. Instead of simply notifying planners about shortages, it would help them understand and resolve the underlying problem.

Simplified customer scenario
Available stock
700 kg
Existing commitments
1,300 kg
Urgent new demand
500 kg

What should the system do?

  1. 01Detect the material constraint and identify affected orders.
  2. 02Evaluate alternative sourcing, substitution, reallocation or delayed confirmation.
  3. 03Compare feasible options by customer impact, cost, delivery commitments and business rules.
  4. 04Present a recommended resolution with explanations and trade-offs.
  5. 05Ask for human approval for consequential changes, then execute authorised actions and monitor the outcome.
DiagramAn illustrative decision from the engagement, simplified. Anonymised enterprise customer, make-to-order manufacturing.

My design contribution. I helped translate the operational problem into an end-to-end agentic execution concept, defining how AI-driven analysis, human decisions and enterprise system actions connect into one coherent experience. The focus wasn't a conversational interface. It was an operational experience that makes autonomous work understandable and actionable.

Key questions we answered

  • What can the system decide on its own?
  • When should it involve a planner?
  • How should it explain competing options?
  • How do users stay in control without supervising every single action?
  • Concept

    A concrete scenario and an end-to-end solution concept

    Ready to prototype and evaluate agentic order fulfilment against realistic operational constraints.

  • Reference

    A reusable reference scenario

    For evaluating broader enterprise AI experience patterns: proactive monitoring, decision support, human approvals and execution transparency.

This describes the discovery and proposed solution, not a production deployment or a verified customer business-impact result.

04

Reflection

From engagements to product direction

Beyond individual customer engagements, the Incubator turns real operational requirements into reusable enterprise AI product direction. My contribution connects three levels of impact.

  • Customer level

    Focused, testable solutions

    Turning complex operational challenges into focused, testable solutions with clear business value.

  • Product level

    Reusable patterns

    Identifying reusable interaction and execution patterns for agentic enterprise software, including delegation, governance, monitoring and human intervention.

  • Organisational level

    A repeatable method

    Establishing a repeatable discovery method and a cross-functional way of working that helps teams make faster, evidence-based decisions about AI opportunities.

The broader ambition is to shift enterprise AI from helping people with individual tasks to reliably delivering meaningful business outcomes within boundaries people define.