
An industrial manufacturer operated complex facilities with hundreds of critical assets and planned their maintenance from manual logs, disconnected systems, and fixed inspection schedules. The work order system knew what had broken and what had been fixed. The sensor feeds knew how each machine was running. Nobody could see an asset's condition, repair history, and parts availability in the same place, so equipment failures caused unplanned downtime, repairs waited on parts, and spare inventory ran high. Mean time between failure varied widely by site and asset class, and the existing ERP and maintenance management tools recorded transactions without predicting anything.
The team built the maintenance system on Palantir Foundry around one ontology. Its core objects are the asset, facility, work order, inspection, sensor stream, part, and bill of materials item. Foundry pipelines ingest sensor readings and maintenance updates continuously and enforce data quality, freshness monitoring, and role-based access, so each asset carries its work order history, its inspections, its parts, and its live readings.
Predictive maintenance models use historical work orders and live sensor data to forecast failure risk and recommend a targeted inspection or part replacement. When vibration and temperature readings cross defined thresholds, the system flags elevated failure risk, recommends an inspection window, and generates a prioritized work order with the parts it needs. Planners and technicians reach those recommendations in Palantir AIP through natural-language prompts and guided actions, and every recommendation is explainable and auditable against safety and compliance requirements.
Planners work in a web application that puts condition data, maintenance history, and upcoming inspections for an asset on one screen. Technicians use the mobile version on site to review recent work, request parts, and receive the recommendation for the machine in front of them.
The system was delivered in eight weeks in sprints. Data integration and ontology design came first and were settled with the maintenance planners and reliability engineers who would use the model. The application build, model development, and pilot user acceptance testing followed. The team settled how each sensor stream maps to its asset before building on it, because a stream tied to the wrong asset sends people to the wrong machine. Maintenance leaders validated the KPIs, and training and a production support model were in place before the rollout widened.
During the pilot phase, unplanned downtime for critical equipment fell by a double-digit percentage. Work orders are triggered earlier and prioritized by risk, so maintenance response is faster, and targeted inspections with timely part replacements increased mean time between failure on high-risk assets. More precise planning reduced unnecessary inspections and emergency repairs. Technicians spend less time searching for information, and a central maintenance history gives every facility the same standard and the same audit trail. The manufacturer is extending the system to more facilities.
Client details in this case study are generalized, and in places combined across engagements, to protect confidentiality. BD Emerson is part of Andersen Consulting, and this engagement was delivered by Andersen Consulting's AI and Advanced Analytics practice, the team we work with on Palantir Foundry and AIP programs. The build reflects how our Palantir Foundry implementation team works in industrial settings, and our guide to what Palantir Foundry is explains the architecture underneath. Teams that need the data layer first can start with our data engineering practice. We are glad to walk through comparable work under NDA.