Reimagining production with Lean PLM - A sovereign approach using Agentic AI

For many manufacturers, PLM serves as the single source of truth for product and process knowledge. It defines how work should be done. Yet on the shop floor, especially during machine setups and changeovers, the reality often looks very different. Operators adapt, search for information, rely on experience, and work around gaps that PLM never sees.

This disconnect is one of the key reasons why the setup time continues to impact availability and reduce overall equipment effectiveness (OEE).

A few years ago, we simply didn’t have the technological maturity or flexibility to address this challenge effectively. When I was consulting for a spindle manufacturing company, setup changes routinely stretched to 10–15 hours. The delays came from a mix of internal and external activities like searching for tools, changing fixtures, interpreting setup instructions, and filling coordination gaps. Everyone understood that setup time was the bottleneck, but the available technology couldn’t capture real execution data, distinguish internal from external work, or reliably feed learning back into PLM. We were working hard, but without the systems, we needed to learn from that work.

Lean PLM aims to close this gap, not by adding more documentation, but by enabling PLM to continuously learn from real execution. In this context, sovereign means keeping manufacturing data, operational intelligence, and AI driven learning loops firmly under the manufacturer’s control, typically on premise or at the edge and tightly integrated with existing industrial systems.

This is where agentic AI plays a truly practical and enabling role. With multiple focused agents supporting setup execution, PLM is no longer disconnected from the shop floor. Instead, it actively assists operators, learns from real setups, and continuously evolves standard work. What was once a manual, experience driven process becomes structured, repeatable, and steadily improving, boosting availability and unlocking meaningful gains in OEE.

In practice, these agents build on the systems that manufacturers already trust. They draw on enablers such as MES integration, machine and controller logs, IoT signals, digital work instructions, and lightweight edge analytics, enhancing existing capabilities rather than replacing them.

"Lean PLM is not a tool upgrade; it’s a learning upgrade."
- Rainer Mewaldt, CTO - PLM, Atos

Insights provided by the setup time

Setup time offers a rare, unfiltered view of shop floor inefficiencies. The machine is stopped, people are fully engaged, and the small sources of waste become visible. Tools are searched for, fixtures are rechecked, instructions are interpreted, and delays quietly accumulate. While PLM often holds most of the required information, it rarely delivers it in a way that supports operators in the moment they need it most.

Agentic AI helps bridge this gap by introducing focused AI agents that actively support PLM during execution, making information timely, contextual, and genuinely useful on the shop floor.

What Lean PLM looks like during a setup

Imagine a machine completing its final batch and the setup phase about to begin. In the past, this moment often introduced uncertainty with operators searching for the right fixtures, double checking program versions, and relying heavily on personal experience to bridge information gaps. While PLM technically held the necessary details, it rarely delivered them in a way that supported operators in the flow of real time execution.

With Lean PLM, the experience looks very different. Before the operator even begins, the PLM link agent automatically pulls the correct operation details, tools, fixtures, and setup instructions directly from PLM, eliminating version confusion and the time spent searching for information. While the machine is still completing its final cycle, the SMED agent identifies what can be prepared in advance, helping teams reduce downtime before it even occurs. (Note SMED is a lean approach to reduce setup time by preparing work in advance and minimizing machine downtime.)

As the changeover begins, the operator guidance agent supports the operator step by step, turning PLM knowledge into clear, timely guidance at the machine instead of leaving it buried in lengthy documents. In the background, the time and event capture agent records setup activities automatically with no stopwatches, observers, or manual data collection required.

Across repeated setups, the root cause agent uncovers recurring delays, and validated improvements flow back into PLM to refine future standards and ensure every step is faster, smarter and more consistent. This is Lean PLM in practice, directly improving availability and strengthening OEE.

Closing the loop: Lean PLM in action

When these agents work together, a true closed loop begins to take shape. PLM establishes the initial standard, the shop floor executes with guided support, real time data reveals where waste occurs, improvements are identified, and PLM is continuously updated with stronger, more accurate standards. Each setup quietly elevates the next one.

At a system level, this forms a closed loop: PLM defines the standard, MES and edge systems capture real execution data, AI agents analyze and contextualize deviations, and validated improvements flow back into PLM. The result is a continuously refined body of standard work that is smarter, sharper, and more aligned with what happens in production.

"Lean PLM acts as a strategic lever to maximize customer and business value across the full product lifecycle, from concept through end of life."
- Mathias Grassmann – Head of PLM, Atos

Brace for change: Getting started

A practical way to begin is by selecting a single pilot line or a critical machine where setup time is already known to be a bottleneck. Start by instrumenting setup activities, supporting operators with guided execution, and capturing real changeover data as it happens. Feed validated improvements back into PLM, then expand the approach line by line. As standards strengthen and learning accelerates, the value compounds, turning each improved setup into a building block for broader operational excellence.

Delivering tangible results By focusing on setup time through a Lean PLM approach, manufacturers can significantly improve availability and unlock 10% to 15% gains in OEE, often without investing in new machines or major capital upgrades. Variability decreases, dependence on individual experience reduces, and operational knowledge stays within the system where it can be shared, strengthened, and continuously improved.

A greenfield opportunity

Despite years of digital investment, much of the shop floor is still a greenfield opportunity for PLM driven execution intelligence.

Agentic AI doesn’t replace PLM or the people who run the operations. It helps PLM learn from real work, capturing what actually happens on the shop floor and turning it into actionable insights.

"Sovereign manufacturing is about closing the loop from Lean E2E PLM through the golden triangle to the edge — ensuring full industrial data control while learning continuously from real-time production."
- Murli Srinivas, Expert Sales – Digital Applications, Atos

Lean PLM succeeds when planning and execution finally speak the same language.

>> Find out how you can accelerate your smart manufacturing transformation journey and modernize operations seamlessly with Atos Industry360 services:
https://atos.net/en/services/digital-applications/industry360