Overview
Most corporate IT asset refresh programmes use a simple rule: replace any laptop older than 4 years. This wastes budget on well-performing machines while leaving struggling assets in service. This project builds a performance-based replacement scoring system using unsupervised ML.
Methodology
Data Collection
Performance telemetry collected from 8,000+ Shell India endpoints via SCCM:
- CPU performance scores over time
- Memory utilisation and error rates
- Disk health (S.M.A.R.T. data)
- Crash frequency and blue-screen events
- Battery health (laptops)
- Application load times
Anomaly Detection (Isolation Forest)
Isolation Forest applied to the full telemetry dataset identifies machines with statistically unusual performance profiles — high crash rates, degraded storage, thermal throttling events.
Clustering (K-Means)
K-Means (k=5) segments the fleet into performance tiers:
- Tier 1: High performers — defer replacement
- Tier 2: Stable — standard lifecycle
- Tier 3: Declining — schedule replacement
- Tier 4: At-risk — priority replacement
- Tier 5: Critical — immediate action
Replacement Priority Score
Combined score (anomaly score × cluster tier weight × user productivity impact) generates a ranked replacement queue submitted monthly to IT procurement.
Results
- 22% reduction in IT support tickets related to hardware issues
- Budget reallocation: £180K saved in Year 1 by deferring replacement of healthy machines
- Replacement queue now data-driven with full audit trail
Delivery
Deployed as a monthly batch job on Azure. Power BI dashboard provides IT managers fleet health visibility.