Novelis runs 32 plants across 9 countries making flat-rolled aluminum.
Last year they connected their predictive maintenance signals directly to their inventory systems across every site.
Most plants have not done that and that is where the downtime happens.
Before we explore into how Novelis closed the gap, quick question:
How is your plant connecting predictive maintenance signals to parts ordering?
👋🏻 I'm Leonardo Ubbiali. This week we're looking at why the most expensive part of unplanned downtime is not the diagnosis, but the parts delay that nobody planned for.
Most predictive maintenance deployments pause at the alert.
The model spots the degradation and generates the recommendation.
A person then checks inventory, orders the part, schedules the technician, and reserves the line slot.
By the time those four things happen, the failure window has closed.
The parts were always the main issue.
Daniel Marchant, Service Manager at Xylem, said it plainly on the MaintainX 2025 panel:
"It typically hasn't been the responsibility of the maintenance and reliability side of the house to think about parts and inventory management to the same extent as the supply chain side of the house."
The alert and the part live in two different systems.
Novelis went enterprise-wide with SymphonyAI's Predictive Asset Intelligence platform in January 2025.
The platform connects predictive signals directly into the maintenance and inventory workflows already running across 32 plants, including the ones with legacy ERPs and 1960s equipment.
What that looks like on an average day at a Novelis plant is that the sensor data flags a bearing degrading on a rolling mill.
The system checks the parts crib, finds the replacement is in stock at the right plant, generates the work ticket, books the technician for the next shift change, and reserves the line slot.
The maintenance lead gets one notification with the plan already built and the shift planner sees the line slot before anyone has to ask.
What Novelis bought from SymphonyAI is the integration into existing workflows, and not just another model sitting on top of their data.
Kruger, the North American tissue and paper producer, runs the same play with the Accenture, Avanade, and Microsoft agentic factory on Azure.
When the system spots degradation, it prepares the maintenance ticket and the spare parts tickets without anyone touching it.
Eric Ashby, Kruger COO, says a 10 to 15 % MTTR reduction is worth 6 figures across their sites.
Verusen, named a 2025 Gartner Cool Vendor for AI-Driven MRO Supply Optimization, shows the working capital side of the same closed loop.
Customers typically give up 20 to 30 % of MRO inventory while cutting stockout risk, by cleaning up master data across fragmented ERPs before automating anything.
Predictive maintenance and parts are one system.
Five things you can do this quarter
The issue: Before deploying predictive maintenance, your plant needs to know whether the data foundation can support it.
What you need: Your current sensor coverage, historian setup, ERP and CMMS systems, MRO master data quality, and the state of your asset hierarchy.
The Prompt (copy this):
I'm a [YOUR ROLE] at a [FACILITY TYPE] manufacturing plant evaluating predictive maintenance. Before deploying any model, I want to audit whether our data foundation is ready.
Current state: Sensor coverage on critical assets is [DESCRIBE]. Historian is [SYSTEM AND AGE]. CMMS and ERP are [SYSTEMS]. MRO master data has [DESCRIBE QUALITY ISSUES, e.g. duplicate SKUs, inconsistent part numbers across sites]. Asset hierarchy is [COMPLETE / PARTIAL / NOT BUILT].
Tell me: What data foundation gaps would halt a predictive-maintenance-to-parts loop from working in our plant? Which of these gaps need to be fixed before any AI vendor walks in? What does a realistic 6 to 12 month data readiness roadmap look like for our setup? What questions should I ask any vendor about their assumptions on our data quality?
What you'll see:
A gap analysis of your current data foundation, a prioritized list of fixes that need to happen before AI deployment, a realistic 6 to 12 month readiness roadmap, and a vendor evaluation checklist focused on data assumptions.
Deloitte Predictive Maintenance Position Paper
The section on the closed-loop architecture from predictive signal to CMMS to ERP to procurement is a good public description of what Novelis and Kruger are building.
It also walks through why lower-maturity PdM systems deliver less than 10 % of the full benefit.
Time to value: 25 minutes
Novelis connected the loop across 32 plants in January. Does your predictive maintenance system know what is in your parts crib?
Hit reply. I read every email.
Leo