Manufacturing plans often collapse not because individual decisions are wrong, but because constraints cascade faster than legacy systems can reason.
Zentio co-founder Immo Polewka explains why spreadsheet-driven planning fails under volatility, and why AI-native systems are required to manage interdependent factory decisions.
1. Looking back at your experience in industrial manufacturing, was there a moment when a plan that looked perfectly correct on paper collapsed in reality due to cascading constraints, and what did that failure reveal about the limits of existing planning systems?
Immo Polewka: We see this constantly: a change in scheduling affects machine utilization, which ripples into inventory levels, which then hits cash flow. The reality is that these decisions are deeply interdependent. Legacy systems based on fragile Excel sheets simply cannot oversee or simulate all these ripple effects in real time. This reveals a fundamental flaw: existing systems are static by design. They treat production as a series of isolated steps rather than a dynamic, cascading environment where every choice has a consequence elsewhere.
2. You just mentioned that Excel sheets are too fragile to handle these ripple effects. Besides that technical failure, what is the most concrete financial or operational damage you have seen when factories rely on “shadow IT” like spreadsheets during volatile times
Immo Polewka: The costs are staggering. Millions are lost to suboptimal planning, reduced productivity, and underutilized capacity. We’ve seen excessive capital tied up in inventory and missed delivery deadlines simply because no single person can calculate every scenario by hand. When volatility hits, these “Excel-based” factories slip; they lose the ability to respond to disruptions with confidence, contributing to the broader trend of European manufacturing losing its leading position.
3. Many companies try to add AI on top of legacy ERP or APS systems rather than rebuilding from scratch. From an architectural standpoint, why do those incremental approaches fail when decisions become deeply interdependent?
Immo Polewka: Incrementalism fails because legacy solutions carry structural technical debt and data silos. They are still struggling to adapt to the cloud, let alone AI. Consequently, legacy solutions pivoting toward AI will be outperformed by AI-native solutions in no time, mirroring the previous shift to cloud-native. To provide real value, you can’t just superficially “throw AI on top” of a product. To gain a real advantage, production planning must be fundamentally built with AI at its core. This integration is essential for structuring and centralizing both operational and machine data. It is the only way to effectively combine mathematical optimization with advanced machine learning, which is necessary to manage specialized workflows.
4. “Agentic AI” is now widely used as a label. In zentio’s case, what precisely qualifies a system as agentic, and how do you reconcile autonomous reasoning with the determinism and auditability required on a factory floor?
Immo Polewka: What makes our system agentic is its utilization of AI agents to unify and centralize all operational, machine performance, and production data. These agents don’t just “suggest”; they enable the system to anticipate capacity needs and respond to breakdowns or shortages with the best available options in real time. We reconcile this with the need for confidence by grounding agentic automation in mathematical optimization. This ensures that while the system “reasons” through agents, the output is a mathematically sound plan decision-makers can trust.
5. In industrial environments, a confident but wrong recommendation can be more dangerous than no recommendation at all. Under what conditions does zentio’s system explicitly refuse to propose a plan, and how is that refusal designed into the system?
Immo Polewka: zentio is built to turn “messy” data into a planning layer that factories can trust. Our focus is on giving decision-makers clarity and confidence. While the system is designed to respond to disruptions, its core is mathematical optimization. If the data is too sparse or the constraints are unsolvable, the goal is to provide transparency so the user understands the trade-offs, rather than offering a “fragile” or suboptimal guess.
6. You combine mathematical optimization with machine learning. Which decisions must remain strictly deterministic, which can be learned probabilistically, and how do you resolve conflicts when those two approaches disagree?
Immo Polewka: Mathematical optimization handles the deterministic side: calculating capacity, shift rules, and hard constraints. Machine learning helps us learn from shopfloor data and simulated scenarios to improve decision quality over time. zentio operates as a unified system that intelligently distinguishes between tasks requiring deterministic precision and those requiring probabilistic learning. By incorporating domain-specific logic, the platform ensures that each specialized workflow is managed by the most effective method. Ultimately, Zentio serves to empower the human planner, who always retains the final decision-making authority.
7. In practice, where do planners most often override zentio’s recommendations today, and how does the system distinguish between a legitimate human insight and resistance driven by habit or lack of trust?
Immo Polewka: Currently, many planners still rely on manual decisions and personal experience. We offer full transparency by enabling simulations of planning scenarios, so planners can see the ripple-effects of gut-based planning in real-time and compare it to alternatives. zentio does provide suggestions for planning scenarios, but ultimately planners decide which scenario they want to apply. We ensure they have all the necessary input to make the optimal decision.
8. By encoding heuristics, qualifications and shift rules into the system, zentio formalizes what was previously informal authority. What decision power does this remove from senior planners, and what organizational friction have you observed as a result?
Immo Polewka: It doesn’t remove power; it elevates the standard of decision-making. Senior planners are currently bogged down by thousands of small operational adjustments that are too complex for one person to oversee. zentio empowers them to move from “firefighting” to strategic advantage. The “friction” is usually just the initial switch from messy spreadsheets to a centralized system, but that evaporates once they see the increased productivity and minimized downtime.
9. As experienced planners retire across Europe, how much of zentio’s value is defensive rather than transformative, and what realistically happens to factories that fail to preserve this operational knowledge in systems?
Immo Polewka: It is both. Defensively, we preserve operational knowledge by structuring messy data that currently lives in planners’ heads or Excel sheets. Transformatively, we allow factories to turn that data into a strategic advantage. Factories that fail to modernize will continue to lose millions to suboptimal planning and underutilized capacity, ultimately slipping further behind global competitors.
10. Energy prices are no longer a background variable in Europe. Should production planning become explicitly energy-aware, and how does zentio handle tradeoffs between energy cost, delivery commitments, and operational stability in real time?
Immo Polewka: Absolutely. Every choice in manufacturing has cascading effects. Our vision is to combine all operational data to allow companies to plan ahead strategically. By optimizing machine settings and minimizing idle time in real time, zentio inherently helps manage energy efficiency. This way, tradeoffs such as energy costs versus delivery deadlines can be simulated by decision-makers with total clarity.
11. Most European factories are brownfield environments with fragmented data and mixed-generation equipment. Without a full digital twin, how far can zentio’s agents still reason effectively, and where does data quality become a hard limit?
Immo Polewka: zentio is specifically built for this reality. We specialize in structuring and centralizing the full range of messy operational, machine, and production data. Our AI-native planning layer is designed to be “trusted” even when the data starts messy. By tying mathematical systems and ML pipelines together with agents, we can reason effectively across shopfloor data to provide clarity, even without a perfect “digital twin”.
12. Looking toward 2030, do you expect the boundary between planning and execution to disappear, and which production decisions should remain permanently human even if full automation becomes technically feasible?
Immo Polewka: By 2030, we expect an AI-native operating model where planning and execution are tightly coupled and continuously inform one another. As systems take on the heavy computational work of real-time optimization, people gain the time, clarity, and freedom to focus on higher-level decisions. Human roles increasingly center on setting strategic direction, defining priorities, managing risk, and making judgment calls where context, accountability, and long-term thinking matter most. Rather than separating planning from execution, this approach allows teams to stay closer to the operation while concentrating on the decisions that truly require human insight and responsibility.
Editor’s Note
This interview argues that modern manufacturing failures stem from static planning tools that cannot adapt to cascading constraints in real time.
