Epiroc’s Christel Füllenbach, Global Operations Manager, and Luba Weissmann, Head of Global AI & Data, share how they moved the Service Copilot beyond technical hype to deliver an AI tool that frontline technicians actually use. By shifting accountability from IT to business leadership, they bypassed the typical hurdles of global deployment.
In the world of industrial service, the leap from a successful AI experiment to a global operational capability is where most projects falter. At Epiroc-a leading global productivity partner for the mining and infrastructure industries-navigating this transition required moving beyond technical hype to focus on a unique partnership between business operations and data architecture. This was not merely a matter of deploying code, but of restructuring how the organization views technological accountability.
As the Business Leader for the Service Technician Copilot, Christel provided the essential “boots on the ground” perspective. Her focus was on the human element: ensuring the tool delivered tangible value for technicians maintaining complex equipment in high-stakes field environments where every minute of downtime is critical. In lockstep, Luba built the enterprise-grade foundations-the data platforms, governance, and delivery models-necessary to transform a fragile lab experiment into a robust, reliable global capability. By aligning frontline operational reality with backend technical excellence, they created a scalable blueprint for industrial AI.
At the upcoming Data Innovation Summit, you’ll be sharing your journey from pilot to global productivity with the Service Copilot. What did the initial collaboration look like before any technology or AI model was actually selected
Christel Füllenbach: We started by understanding the reality of field service, where technicians lose time, where decisions carry risk, and where information is hardest to access. The early collaboration was entirely business‑ and user‑driven, long before we discussed AI or platforms.
Luba Weissmann: From the AI & Data side, our focus early on was to avoid the classic mistake of starting with tools. We worked with the service organization to translate operational challenges into structured data and AI use cases. Only once the value and decision flows were clear did we evaluate which technologies could support them. 
Building on that, how did you ensure the business requirements-rather than the technical capabilities-remained the primary driver during the design phase?
Christel Füllenbach: Every design decision was tied to clear business outcomes such as faster diagnostics, reduced onboarding time, and safer service execution. Technicians consistently find the guided diagnostic flow most valuable because it reduces troubleshooting time and provides machine‑specific, experience‑based recommendations.
We prioritized this over more advanced AI features because it solved the biggest operational bottleneck identified in the field and created immediate value in uptime and service efficiency. If a feature didn’t support those outcomes, it didn’t move forward, regardless of how advanced the technology was.
Luba Weissmann: We structured the project so the business defined the problem and success metrics, while the AI & Data team focused on architecture, models, and responsible deployment. That separation of responsibilities ensured technology served the operational goal rather than becoming the goal itself.
How do you manage the tension between the rapid, iterative pace of AI development and the practical, high-stakes reality of global field operations?
Christel Füllenbach: We iterate quickly in controlled environments but deploy cautiously. In field operations, trust and reliability matter more than speed, so only proven and stable improvements reach technicians.
Luba Weissmann: This is where platform and governance matter. We separate experimentation from production. Models can evolve rapidly in controlled environments, but production deployments go through strict reliability, safety, and data quality controls-particularly important when AI is supporting technicians working on critical equipment.
One important guardrail is that the Copilot does not invent new procedures. All recommendations are constrained to validated service knowledge and existing operational documentation. In addition, the system is monitored through feedback loops and usage data so that incorrect or unclear suggestions can be quickly corrected. This ensures that the AI acts as a structured assistant that helps technicians navigate complex documentation faster, rather than replacing established safety and service standards.
Let me know if anything should be adjusted further.
In many projects, the business “requests” a tool and IT “delivers” it. How did your approach change when the Business Owner became the one actually accountable for the AI’s performance?
Christel Füllenbach: The Copilot became an operational capability, not an experiment. With business ownership came KPIs, governance, and accountability for real‑world impact, not just delivery.
Luba Weissmann: It changes the conversation completely. Instead of a technology delivery, the AI becomes part of the operational system. Our role in AI & Data is to provide the platform, lifecycle management, and governance-but the business remains accountable for value creation and adoption.
Trust is a major barrier for frontline technicians. How did this co-creation model help you win over users who rely heavily on their own intuition and experience?
Christel Füllenbach: By involving technicians early and positioning the Copilot as a support tool -not a replacement-we respected their expertise. That co‑creation built trust and drove adoption naturally.
Luba Weissmann: In practice this grounding was critical. Experienced technicians are understandably skeptical of AI suggestions that cannot be verified. By linking the Copilot directly to validated service manuals, troubleshooting procedures, and internal knowledge bases, technicians can immediately see where the recommendation comes from. This allows them to cross-check the suggestion against the same documentation they already trust, which significantly reduced the initial resistance and made the Copilot feel like an extension of their existing expertise rather than a “black box”.
Now that the Service Copilot is scaling globally, how do you ensure the AI remains aligned with the shifting priorities of the business?
Christel Füllenbach: Continuous feedback, usage data, and business governance ensure the Copilot evolves in line with service strategy and operational priorities across regions.
Luba Weissmann: AI systems are not static products. We treat them as continuously evolving capabilities supported by data pipelines, monitoring, and feedback loops. That allows the Copilot to improve as new service knowledge, machine data, and operational insights emerge.
For those stuck in “pilot purgatory,” what is the one non-technical factor that allowed you to successfully transition to a global-scale operation?
Christel Füllenbach: Business ownership transformed the rollout from a pilot phase into a governed operational capability. With the business accountable for KPIs, adoption, and global consistency, the rollout became structured, faster, and fully aligned with service priorities. IT remained essential, but the business ensured deployment happened only once the solution was stable and trusted in real field environments.
Luba Weissmann: Most pilots fail to scale because the organization lacks the foundations around them. Scaling AI requires three things simultaneously: clear business ownership, a production-grade data and AI platform, and governance that allows innovation without losing operational control.
Join Christel Füllenbach and Luba Weissmann at the Data Innovation Summit for their session, “AI in field service: from pilot to global productivity gains.” This deep dive explores the specific strategies used to measure productivity and navigate the complexities of a global rollout in a regulated industrial setting. Gain first-hand insights into the practical decisions and lessons learned that turn early AI concepts into high-impact, deployed reality.
