Predictive Parts Lifecycle Intelligence for Fleet Operations
Built a parts lifecycle management system that digitized inventory tracking, predictive ordering, and maintenance scheduling across 2,000 vehicles — reducing component failures by 18% and transforming reactive parts management into a data-driven, predictive operation.
Shadowed engineering teams, analyzed failure data, and reviewed procurement logs to uncover four systemic pain points.
Parts were replaced only after failure — leading to emergency procurement at premium costs, extended vehicle downtime, and cascading route disruptions. No system existed to predict end-of-life.
Each depot tracked inventory independently using spreadsheets and paper logs. Engineering had no fleet-wide view of which components were installed where or how many service hours remained.
Procurement ordered based on historical averages and gut feel — causing simultaneous stockouts of critical parts (brake pads, engine belts) and overstocking of slow-movers that tied up capital.
Service records, inventory, maintenance schedules, and procurement lived in separate systems. Engineers couldn't answer "Which buses need brake pads in 30 days?" without cross-referencing spreadsheets.
Embedded with depot teams for four weeks. Observed workflows, interviewed stakeholders, analyzed procurement and failure data.
Frontline technician managing 80-120 vehicles per depot. Relies on paper forms and verbal handoffs. Spends 30%+ of time chasing parts availability.
Manages 50+ suppliers across mechanical, electrical, and body components. Forecasts demand using last year's volumes plus a gut-feel buffer.
Responsible for daily fleet readiness. A single parts delay cascades into route coverage gaps affecting thousands of commuters.
A parts lifecycle layer on top of existing tools — faster to deploy, lower risk, and maintenance teams kept familiar workflows.
Cataloging every sub-component would have delayed launch by 12+ months. Started with major assemblies, planned to decompose later.
Started with the 10 highest-failure-rate components (brake pads, engine belts, air compressors) that drove 70% of unplanned maintenance.
Kept humans in the loop for purchase decisions during phase one — trust needed to be built before automating spend.
Risk-weighted prioritization: failure frequency × downtime cost × procurement lead time. 10 critical component categories covering 70% of unplanned events.
Every part installation recorded against a specific vehicle ID: part number, supplier, installation date, service hours at install, expected lifecycle hours. This was the data backbone — without it, the system was just another inventory tracker.
Usage-based lifecycle models for each of the top 10 component categories, built from 3 years of historical failure data. Each model predicted failure probability based on service hours elapsed — enabling "replace by X date" forecasts.
Automated reorder calculations combining: remaining lifecycle hours across the fleet, supplier lead time, and safety stock thresholds. Generated pre-order recommendations with quantity, urgency, and optimal order date.
Real-time depot view: vehicles service-ready (green), in maintenance (yellow), blocked on parts (red). Drill into any vehicle for component status, upcoming replacements, and parts availability.
Every major component tracked against a specific bus with installation date, service hours, and lifecycle status — the foundation for all future asset management.
Depot teams moved from "fix what broke today" to "plan what needs replacing this month" — a fundamental change in how the engineering function operated.
The 10-component pilot proved the model worked. KMB could extend to all 5,000+ part types using the same architecture and lifecycle methodology.
Demand forecasting data enabled longer-term supplier contracts at lower unit costs — replacing spot-buying at emergency rates.
The hardest part wasn't building the system — it was earning trust from maintenance engineers who'd managed parts with paper and intuition for decades. Four weeks embedded in depots, watching how they worked, before designing anything. The system had to fit their workflow, not the other way around.
Cataloging all 5,000+ part types would have been thorough — and would have killed the project. 10 highest-failure-rate components gave us 70% of the impact with 2% of the catalog complexity. Not all components are equal.
Every month the system ran, the models got more accurate — more data points, better failure curves, tighter pre-order windows. This wasn't just solving today's problem; it was building an asset that gets more valuable over time.
Involve procurement earlier. We built the pre-order engine from engineering requirements (lifecycle hours, failure probability) but didn't fully account for procurement realities (minimum order quantities, volume discounts). Bringing procurement in at the design stage would have produced better recommendations from day one.