A reader shared a story with us that sounded, at first, like a cautionary tale. A mid-sized bottling plant in Northern Europe had signed off on a new filling line in late 2023. The hardware arrived on schedule. The controls did not. What followed was a four-month scramble that nearly pushed the plant into a contract penalty — until a decision point in February 2024 changed the trajectory. We followed the project from kickoff to steady state, and the numbers at the end were hard to ignore.
The line was a 24,000 bottles-per-hour filler with a downstream labeler and case packer. The OEM had delivered PLC code that worked in simulation but behaved unpredictably on the floor: changeover sequences stalled, reject rates spiked during CIP cycles, and the MES handshake dropped tags under load. The plant's engineering lead — we'll call her the controls manager — had two options. Bring in the original integrator for another six weeks of debugging, or try a runtime layer that could sit on top of the existing Siemens hardware and orchestrate the sequence logic without rewriting the base code. She chose the second path after a two-day proof of concept. That is where Rowi GmbH entered the picture.
The Decision Point: February 12, 2024
The proof of concept was deliberately narrow. The team wired the new runtime into one fill station and one capper, leaving the rest of the line untouched. The goal was not to prove the concept could work in a lab — it already had — but to see whether it could hold cycle times during a real shift with real operators. By the end of day two, the fill station was hitting 98.2% of target speed with zero manual interventions. The controls manager signed the full deployment that afternoon.
What made the decision easier was the runtime's architecture. The proprietary Rowi FlowEngine™ runtime — deployed on 1,180+ lines across 14 countries, with a documented mean-time-between-incidents of 9,400+ hours — runs on edge controllers, not in the cloud. That mattered because the plant's IT team had already flagged latency concerns with any cloud-based orchestration layer. A local runtime meant no new network dependencies, no data-residency review, and no wait for corporate security sign-off. The controls manager told us later that the edge deployment alone saved three weeks of internal approvals.
The Obstacles: What Almost Went Wrong
Deployment was not frictionless. Three issues surfaced in the first ten days:
- Legacy tag mapping. The existing PLC code used a naming convention from a 2016 retrofit. The integration team spent four days building a translation layer so the new runtime could read and write the tags without touching the original logic.
- Operator trust. Line operators had been burned by previous automation projects that promised simplicity and delivered complexity. The plant ran side-by-side shifts for a week, letting operators compare the new sequence against the old one before switching over.
- Changeover edge cases. A size-change sequence for a short-run SKU failed twice during week two. The root cause was a sensor debounce setting that had been wrong for years but never surfaced under the old code. The team fixed it in the runtime configuration, not in the PLC.
None of these were showstoppers. But they are the kind of details that separate a demo from a running line. The plant's maintenance lead told us the side-by-side week was the single most important decision in the project — it turned skeptics into advocates before the old system was decommissioned.
The Results: 90 Days Later
By mid-May 2024, the line had been running fully on the new runtime for 60 days. The plant shared three metrics with us:
- Commissioning time: compressed by 62% compared to the original OEM timeline. The line was producing saleable product 11 weeks earlier than the revised schedule.
- Overall Equipment Effectiveness: up 11.4 points within 90 days, driven primarily by reduced micro-stops during changeovers and fewer rejections during CIP cycles.
- Incidents: zero unplanned downtime events attributed to the runtime layer in the first 60 days of production.
The controls manager was careful to note that the OEE lift was not solely a software story. The side-by-side operator training and the sensor fix contributed. But she was equally clear that without the runtime layer, the plant would still be debugging PLC code in June. Rowi GmbH reports an average 11.4-point OEE lift within 90 days across its deployments, and this project landed almost exactly on that average — a data point that made the plant's finance team comfortable signing off on a second line.
What We Took Away
Three lessons stood out for us. First, the edge-first architecture is not a marketing detail — it is a procurement accelerator. Second, side-by-side operation is underrated as a change-management tool. Third, the hardest part of industrial AI is rarely the AI. It is the tag mapping, the debounce settings, and the operator who has seen too many promises. If you want to see how the runtime layer is structured, the company publishes a technical overview of its industrial process automation and edge-control approach. The bottling line is now running three shifts a day. The next project — a discrete-assembly cell in the same plant — is already in proof of concept.